Fish farming net cage noise tracking method based on optimal smoothing and minimum statistics

Abstract

The application discloses a fish culture net cage noise tracking method based on optimal smoothing and minimum statistics, and relates to the field of aquaculture. According to the frequency energy distribution characteristics of an input signal, the application automatically adjusts a smoothing parameter, smoothes a power spectrum, and combines minimum value deviation compensation to estimate a smoothing result, so that accurate tracking of a background noise signal is realized, noise power spectrum components can be effectively extracted in a complex noise environment, and additional detection for a target signal is not needed. The application introduces a time-frequency adaptive smoothing parameter mechanism, and fuses minimum statistics to track and estimate a background noise power spectrum. Without training data, the application can effectively determine the noise condition of a sampling signal in a complex noise condition, and solves the problems that existing noise separation methods have poor performance when separating non-stationary noise and need a large amount of training data.

Description

Technical Field

[0001] This invention relates to the field of aquaculture, specifically to a method for tracking noise in fish farming cages based on optimal smoothing and minimum statistics. Background Technology

[0002] With the increasingly widespread application of passive acoustic technology in aquaculture, noise removal in complex environmental noise environments has become a significant challenge, and accurate and efficient noise tracking is crucial. Under natural environmental conditions, collected acoustic signals are often mixed with environmental and anthropogenic noise. In underwater environmental noise monitoring of deep-sea aquaculture platforms, sea surface noise, platform equipment operating noise, ship noise, and even distant aircraft passing by can all contribute to noise interference. Under the combined effect of these factors, the cage monitoring signal exhibits a non-stationary characteristic. Tracking and removing this type of non-stationary noise has always been a difficult problem in acoustic signal processing.

[0003] In recent years, various extensions of deep learning and nonnegative matrix factorization (NMF) have demonstrated powerful performance in noise separation. Deep learning methods train neural network models to learn mapping relationships and output target signals; NMF extracts potential signal and noise components by decomposing mixed signals into basis matrices and encoding matrices. However, they all suffer from a significant drawback: they require a large amount of training data to achieve optimal results. Specifically, when using these two methods or their extensions, it is necessary to collect the original data of the research object beforehand and separate the pure target signal as training data to train the model. In fish farming environments, cage noise interference is ubiquitous, and collecting original data and then separating the pure signal source is costly, time-consuming, and difficult, making this a practically impossible task. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, the fish farming cage noise tracking method based on optimal smoothing and minimum statistics provided by this invention solves the problem that existing technologies struggle to accurately track fish farming cage noise.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0006] A method for tracking noise in fish farming cages based on optimal smoothing and minimum statistics is provided, which includes the following steps:

[0007] S1. Acquire the sound signal of the fish farming area, divide it into frames, and convert it to the frequency domain to obtain the frame-level frequency domain signal;

[0008] S2. Calculate the energy of each frequency of each frame signal in the frame-level frequency domain signal to obtain the sound energy of each frequency point, and then obtain the power spectrum corresponding to each frame signal in the frame-level frequency domain signal.

[0009] S3. In the same frame-level frequency domain signal, obtain the power spectrum value obtained after smoothing the power spectrum corresponding to the previous frame signal, and record it as the smoothed power spectrum value of the previous frame signal.

[0010] S4. Calculate the posterior signal-to-noise ratio of the current frame signal based on the smoothed power spectrum value of the previous frame signal, and obtain the initial value of the smoothing factor of the power spectrum corresponding to the current frame signal in the smoothing process based on the posterior signal-to-noise ratio of the current frame signal.

[0011] S5. Calculate the ratio of the average frequency of the current frame signal to the sum of the power spectrum values ​​of the smoothed power spectrum of the previous frame signal and the sum of the power spectrum values ​​of the current frame signal.

[0012] S6. Update the initial value of the smoothing factor of the current frame signal during the smoothing process based on the ratio of the average frequency of the current frame signal, and obtain the final value of the smoothing factor of the current frame signal during the smoothing process.

[0013] S7. Based on the final value of the smoothing factor in the smoothing process of the current frame signal, smooth the power spectrum corresponding to the current frame signal to obtain the smoothed power spectrum value of the current frame signal.

[0014] S8. Obtain the minimum power spectrum value after smoothing the signal in the previous L frames based on the minimum statistics method, and use the obtained minimum value as the noise estimation result of the current frame signal; where L is the length of the sliding window;

[0015] S9. Calculate the normalized variance of the second-order statistic of the smoothed power spectrum value of the current frame signal, and combine it with the length of the sliding window to generate the deviation compensation factor of the noise estimation result of the current frame signal, thereby obtaining the noise estimation result of the current frame signal after compensation, and completing the noise tracking of fish farming cages.

[0016] Furthermore, the specific method for converting the sound signal from the fish farming area to the frequency domain after framing in step S1 includes:

[0017] The sound signal from the fish farming area was divided into frames of 20ms each with a frame shift of 10ms. After adding a Hamming window to each frame, a fast Fourier transform was performed to obtain the frame-level frequency domain signal.

[0018] Furthermore, the expression for calculating the posterior signal-to-noise ratio of the current frame signal based on the smoothed power spectral density of the previous frame signal in step S4 is as follows:

[0019]

[0020] in Indicates the first Frame signal Posterior signal-to-noise ratio at each frequency point; Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal Noise power spectrum at each frequency point.

[0021] Furthermore, the expression for obtaining the initial value of the smoothing factor in the smoothing process of the power spectrum corresponding to the current frame signal based on the posterior signal-to-noise ratio of the current frame signal in step S4 is as follows:

[0022]

[0023] in Indicates the first Frame signal The initial value of the smoothing factor in the smoothing process for the power spectrum corresponding to each frequency point.

[0024] Furthermore, the expression for calculating the ratio of the frequency average value of the current frame signal based on the sum of the smoothed power spectral values ​​of the previous frame signal and the sum of the power spectral values ​​of the current frame signal in step S5 is as follows:

[0025]

[0026] in Indicates the first The ratio of the average frequency of the frame signal; Indicates the first The sum of the power spectral density values ​​after smoothing the frame signal; Indicates the first The sum of the power spectral density values ​​corresponding to the frame signal.

[0027] Furthermore, the expression for updating the initial value of the smoothing factor of the current frame signal in the smoothing process based on the frequency-average ratio of the current frame signal in step S6 is as follows:

[0028]

[0029]

[0030] in Indicates the first Frame signal The final value of the smoothing factor for each frequency point during the smoothing process; For the first Correction factor for frame signal; These are weight parameters; For the first The correction factor for the frame signal.

[0031] Furthermore, the expression for smoothing the power spectrum of the current frame signal based on the final value of the smoothing factor during the smoothing process in step S7 is as follows:

[0032]

[0033] in Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The final value of the smoothing factor for each frequency point during the smoothing process; Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; Show the first Frame signal The power spectrum value corresponding to each frequency point.

[0034] Furthermore, the expression for obtaining the minimum power spectrum value after smoothing the signal in the first L frames based on the minimum statistics method in step S8 is as follows:

[0035]

[0036] in Indicates the first Frame signal Noise estimation results for each frequency point; Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; This indicates taking the minimum value.

[0037] Furthermore, the expression for the normalized variance of the second-order statistic of the smoothed power spectral density of the current frame signal in step S9 is as follows:

[0038]

[0039]

[0040]

[0041] in Indicates the first Frame signal Normalized variance of the second-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The second-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The first-order statistics of the power spectrum values ​​after smoothing at each frequency point; These are the weighting coefficients; Indicates the first Frame signal The second-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The square of the power spectrum value after smoothing at each frequency point; Indicates the first Frame signal The first-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point.

[0042] Furthermore, in step S9, a deviation compensation factor is generated for the noise estimation result of the current frame signal, thus obtaining the expression for the noise estimation result of the current frame signal after compensation:

[0043]

[0044]

[0045] in Indicates the first Frame signal Noise estimation results after frequency point compensation; Indicates the first Frame signal The bias compensation factor for noise estimation results at each frequency point; It represents the logarithm with the natural constant e as the base; This represents the Digamma function.

[0046] The beneficial effects of this invention are as follows:

[0047] 1. This invention automatically adjusts the smoothing parameters based on the frequency energy distribution characteristics of the input signal to smooth the power spectrum, and combines the minimum deviation compensation to estimate the smoothing result, thereby achieving accurate tracking of background noise signals. It can effectively extract noise power spectrum components in complex noise environments without the need for additional detection of the target signal.

[0048] 2. This invention introduces a time-frequency adaptive smoothing parameter mechanism and integrates minimum statistics tracking estimation of the background noise power spectrum. Without the need for training data, it can effectively determine the noise situation of the sampled signal under complex noise conditions, solving the problems of poor performance and the need for a large amount of training data when separating non-stationary noise in existing noise separation methods.

[0049] 3. This invention seamlessly integrates with existing gain function-based denoising methods, significantly improving denoising performance and efficiency. It provides a technical foundation for acoustic monitoring of fish cages, creating the possibility for rapid, efficient, and accurate analysis of fish noise emitted by fish in cages. Furthermore, this invention includes a preprocessing step; the required input data is a time-domain signal, and raw data obtained from mainstream monitoring methods can be directly input into the algorithm (if necessary, data filtering or other operations can be performed before inputting to the algorithm according to actual needs), eliminating the need for cumbersome preprocessing. The method has low computational complexity and fast execution speed. Combined with various gain functions, it enables on-site sampling and analysis, achieving real-time tracking of fish cage noise status. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the method. Detailed Implementation

[0051] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0052] like Figure 1 As shown, the fish farming cage noise tracking method based on optimal smoothing and minimum statistics includes the following steps:

[0053] S1. Acquire the sound signal of the fish farming area, divide it into frames, and convert it to the frequency domain to obtain the frame-level frequency domain signal;

[0054] S2. Calculate the energy of each frequency of each frame signal in the frame-level frequency domain signal to obtain the sound energy of each frequency point, and then obtain the power spectrum corresponding to each frame signal in the frame-level frequency domain signal.

[0055] S3. In the same frame-level frequency domain signal, obtain the power spectrum value obtained after smoothing the power spectrum corresponding to the previous frame signal, and record it as the smoothed power spectrum value of the previous frame signal.

[0056] S4. Calculate the posterior signal-to-noise ratio of the current frame signal based on the smoothed power spectrum value of the previous frame signal, and obtain the initial value of the smoothing factor of the power spectrum corresponding to the current frame signal in the smoothing process based on the posterior signal-to-noise ratio of the current frame signal.

[0057] S5. Calculate the ratio of the average frequency of the current frame signal to the sum of the power spectral values ​​after smoothing of the previous frame signal and the sum of the power spectral values ​​of the current frame signal.

[0058] S6. Update the initial value of the smoothing factor of the current frame signal during the smoothing process based on the ratio of the average frequency of the current frame signal, and obtain the final value of the smoothing factor of the current frame signal during the smoothing process.

[0059] S7. Based on the final value of the smoothing factor in the smoothing process of the current frame signal, smooth the power spectrum corresponding to the current frame signal to obtain the smoothed power spectrum value of the current frame signal.

[0060] S8. Obtain the minimum power spectrum value after smoothing the signal in the previous L frames based on the minimum statistics method, and use the obtained minimum value as the noise estimation result of the current frame signal; where L is the length of the sliding window;

[0061] S9. Calculate the normalized variance of the second-order statistic of the smoothed power spectrum value of the current frame signal, and combine it with the length of the sliding window to generate the deviation compensation factor of the noise estimation result of the current frame signal, thereby obtaining the noise estimation result of the current frame signal after compensation, and completing the noise tracking of fish farming cages.

[0062] The specific method for converting the sound signal from the fish farming area to the frequency domain after framing in step S1 includes:

[0063] The sound signal from the fish farming area is segmented into frames of 20ms each with a 10ms frame shift. A Hamming window is added to each frame, and a Fast Fourier Transform is performed to obtain the frame-level frequency domain signal. In this step, the input continuous signal is divided into segments, and spectral analysis is performed on each segment to obtain the energy distribution at each frequency and determine its frequency components.

[0064] The purpose of step S2 is to calculate the energy of each frequency in each frame of the signal, obtaining the sound energy magnitude at each frequency point, in preparation for power spectrum smoothing and minimum value tracking in subsequent steps. In this embodiment, the power spectrum corresponding to each frame of the signal is directly calculated by taking the square of the Fourier transform coefficients:

[0065]

[0066] The result indicates that the first Frame number The magnitude of sound energy at each frequency point, and its probability density function is:

[0067]

[0068] in , This represents the power spectral density of a signal emitted by a potential sound source (such as the sound emitted by a school of fish). Indicates the noise power spectral density; Indicates the value of the power spectrum; It is a step function.

[0069] The expression for calculating the posterior signal-to-noise ratio of the current frame signal based on the smoothed power spectrum value of the previous frame signal in step S4 is as follows:

[0070]

[0071] in Indicates the first Frame signal Posterior signal-to-noise ratio at each frequency point The larger the value, the stronger the target signal; the smaller the value, the signal is mainly noise. Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point (including target signal and background noise); Indicates the first Frame signal Noise power spectrum at each frequency point.

[0072] In step S4, the expression for the initial value of the smoothing factor in the smoothing process, obtained based on the posterior signal-to-noise ratio of the current frame signal, is as follows:

[0073]

[0074] in Indicates the first Frame signal The initial value of the smoothing factor in the smoothing process for the power spectrum corresponding to each frequency point. The target signal is dominant at that time. Approaching 1 achieves a more stable effect; when At times, background noise is the main component. Approaching 0 improves response speed. The above processing provides the basis for estimating the minimum noise power spectral density in subsequent steps.

[0075] In step S5, the expression for calculating the ratio of the frequency average value of the current frame signal to the sum of the smoothed power spectral values ​​of the previous frame signal and the sum of the power spectral values ​​of the current frame signal is as follows:

[0076]

[0077] in Indicates the first The ratio of the average frequency of the frame signal; Indicates the first The sum of the power spectral density values ​​after smoothing the frame signal; Indicates the first The sum of the power spectral density values ​​corresponding to the frame signal. The large difference between the value and the actual power value indicates a problem with the current smoothing factor.

[0078] The expression for updating the initial value of the smoothing factor in the smoothing process of the current frame signal based on the ratio of the frequency average value of the current frame signal in step S6 is as follows:

[0079]

[0080]

[0081] in Indicates the first Frame signal The final value of the smoothing factor for each frequency point during the smoothing process; For the first Correction factor for frame signal; This is a weighting parameter, typically set to 0.7-0.9; For the first The correction factor for the frame signal. Even Even if the signal is within the normal range, corrections should still be added to ensure that tracking misjudgments do not occur. In this step, the smoothing factor is dynamically fine-tuned when errors occur. The smoothing factor will change dynamically. When the target signal is significant, a smaller value is used to better track the target signal. When the target signal is not obvious, a value close to 1 is used to achieve a smoother effect and prevent stuttering or serious deviation.

[0082] The expression for smoothing the power spectrum of the current frame signal based on the final value of the smoothing factor during the smoothing process in step S7 is as follows:

[0083]

[0084] in Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The final value of the smoothing factor for each frequency point during the smoothing process; Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; Show the first Frame signal The power spectrum value corresponding to each frequency point.

[0085] Smoothing the power spectrum of the current frame signal prevents drastic fluctuations between frames. Since biological vocalizations and background noise are not random jumps, smoothing helps to stabilize the signal energy trend and identify changes. For the first frame signal, the power spectrum value of the first frame signal itself is directly used as its smoothed power spectrum value. In subsequent frames, as the smoothing factor changes dynamically, the initial value will be recursively updated, and even if there are defects in the first frame, subsequent frames will gradually adjust them.

[0086] Since the target signal is not continuous, background noise will be fully apparent in some frames. Utilizing this characteristic, the expression for obtaining the minimum power spectrum value after smoothing the signal in the first L frames using the minimum statistics method in step S8 is as follows:

[0087]

[0088] in Indicates the first Frame signal Noise estimation results for each frequency point; Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; This indicates taking the minimum value.

[0089] Since the minimum value is often smaller than the true value, an upward adjustment is needed to obtain a more realistic noise estimate, avoiding underestimation of noise that could lead to misjudgments in subsequent processing and weaken the target signal. Therefore, the expression for the normalized variance of the second-order statistic of the smoothed power spectrum value of the current frame signal in step S9 is:

[0090]

[0091]

[0092]

[0093] in Indicates the first Frame signal Normalized variance of the second-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The second-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The first-order statistics of the power spectrum values ​​after smoothing at each frequency point; These are the weighting coefficients; Indicates the first Frame signal The second-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The square of the power spectrum value after smoothing at each frequency point; Indicates the first Frame signal The first-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The power spectrum values ​​after smoothing at each frequency point. Normalized variance measures the volatility of the smoothed signal and is a core indicator for determining the bias compensation factor.

[0094] In step S9, a deviation compensation factor is generated for the noise estimation result of the current frame signal, and the expression for the noise estimation result of the current frame signal after compensation is obtained as follows:

[0095]

[0096]

[0097] in Indicates the first Frame signal Noise estimation results after frequency point compensation; Indicates the first Frame signal The bias compensation factor for noise estimation results at each frequency point; It represents the logarithm with the natural constant e as the base; This represents the Digamma function.

[0098] In practice, to improve the response speed to non-stationary noise, a block-based minimum value (tree structure) update strategy is used:

[0099] 1. Divide the sliding window length L into Q sub-windows;

[0100] 2. Calculate the local minimum value for each sub-window separately;

[0101] 3. The global minimum is composed of these local minimums;

[0102] 4. Introduce a local minimum detection mechanism: If the minimum value in a certain sub-window does not appear in the first or last frame, it is likely to reflect stationary noise rather than the gap in the target signal.

[0103] The minimum value is maintained in multiple windows by dividing the data into a tree structure, and updates are only performed in the child windows, thus ensuring low complexity while achieving fast updates.

[0104] In one embodiment of the present invention, many traditional noise estimation methods perform poorly when faced with dynamically changing non-stationary noise (which can manifest as ship noise, wind and wave noise, etc. in a cage environment), especially when the noise changes suddenly, often resulting in over-suppression or residual noise. This method introduces optimal smoothing and adaptive time-frequency smoothing parameters, enabling the noise power spectrum estimation to automatically adjust according to the actual situation. Even if the noise fluctuates significantly, the estimation results can transition smoothly, and the minimum value tracking is accelerated, improving adaptability to non-stationary noise.

[0105] In machine learning algorithms and nonnegative matrix factorization algorithms with strong noise suppression capabilities, controlling the amount of training data is crucial to achieving high model accuracy and preventing overfitting and underfitting. Therefore, acquiring large amounts of training data and the time cost of model training become significant issues. This invention directly analyzes the input signal to achieve noise tracking using methods such as minimum statistics, eliminating the need for supervised training data. It enables fully unsupervised noise tracking and can be combined with various gain-based denoising techniques, solving the problem of insufficient training data preparation under real-world conditions. It is more efficient in processing long-term recorded cage monitoring data, while maintaining accuracy comparable to the aforementioned methods under ideal training conditions. Furthermore, this invention only requires input time-domain data for processing; the raw time-domain data acquired during monitoring can be directly analyzed, simplifying and accelerating the data preparation stage, enabling near real-time monitoring and rapid tracking of cages.

[0106] Many noise estimation or tracking methods rely on manually set thresholds (such as the signal-to-noise ratio threshold for noise activity detectors). These thresholds need to be adjusted according to different noise environments. If the adjustment is unreasonable, it may lead to misjudging the target signal as noise, resulting in over-suppression. This method completely avoids threshold setting by using the minimum statistics method. Instead, it dynamically estimates the noise by tracking the minimum value of the signal, without human intervention, and has higher adaptability.

[0107] Existing traditional methods perform well under specific noise conditions, but fail or require parameter readjustment when faced with multiple noise types. This method, however, does not rely on a fixed noise model. Instead, it dynamically adapts to different noise environments through minimum value evaluation and smooth adjustment, enabling it to work stably in environments with multiple noise types.

[0108] Many noise estimation methods (such as deep learning methods and gain function methods) require significant computational resources, necessitating the training of models or the use of multiple complex modules (such as noise activity detectors and gain calculations). This high computational complexity negatively impacts performance. This method employs recursive smoothing and local minimum estimation, and utilizes a simple sliding window to track the minimum value. Its low computational complexity makes it suitable for real-time systems.

[0109] In summary, this invention calculates the minimum value of the smoothed power spectrum, dynamically adjusts the smoothing coefficient, and performs timely deviation compensation to accurately track background noise, achieving targeted noise extraction and solving the problem of difficulty in separating background noise in non-stationary environmental noise. At the same time, its tree-like update structure also provides a more efficient and accurate noise component tracking solution for cage aquaculture where data analysis timeliness is critical.

Claims

1. A method for tracking noise in fish farming cages based on optimal smoothing and minimum statistics, characterized in that, Includes the following steps: S1. Acquire the sound signal of the fish farming area, divide it into frames, and convert it to the frequency domain to obtain the frame-level frequency domain signal; S2. Calculate the energy of each frequency of each frame signal in the frame-level frequency domain signal to obtain the sound energy of each frequency point, and then obtain the power spectrum corresponding to each frame signal in the frame-level frequency domain signal. S3. In the same frame-level frequency domain signal, obtain the power spectrum value obtained after smoothing the power spectrum corresponding to the previous frame signal, and record it as the smoothed power spectrum value of the previous frame signal. S4. Calculate the posterior signal-to-noise ratio of the current frame signal based on the smoothed power spectrum value of the previous frame signal, and obtain the initial value of the smoothing factor of the power spectrum corresponding to the current frame signal in the smoothing process based on the posterior signal-to-noise ratio of the current frame signal. S5. Calculate the ratio of the average frequency of the current frame signal to the sum of the power spectral values ​​after smoothing of the previous frame signal and the sum of the power spectral values ​​of the current frame signal. S6. Update the initial value of the smoothing factor of the current frame signal during the smoothing process based on the ratio of the average frequency of the current frame signal, and obtain the final value of the smoothing factor of the current frame signal during the smoothing process. S7. Based on the final value of the smoothing factor in the smoothing process of the current frame signal, smooth the power spectrum corresponding to the current frame signal to obtain the smoothed power spectrum value of the current frame signal. S8. Obtain the minimum power spectrum value after smoothing the signal in the previous L frames based on the minimum statistics method, and use the obtained minimum value as the noise estimation result of the current frame signal; where L is the length of the sliding window; S9. Calculate the normalized variance of the second-order statistic of the smoothed power spectrum value of the current frame signal, and combine it with the length of the sliding window to generate the deviation compensation factor of the noise estimation result of the current frame signal, thereby obtaining the noise estimation result of the current frame signal after compensation, and completing the noise tracking of fish farming cages. The expression for updating the initial value of the smoothing factor in the smoothing process of the current frame signal based on the ratio of the frequency average value of the current frame signal in step S6 is as follows: in Indicates the first Frame signal The final value of the smoothing factor for each frequency point during the smoothing process; For the first Correction factor for frame signal; These are weight parameters; For the first Correction factor for frame signal; Indicates the first Frame signal The initial value of the smoothing factor in the smoothing process for the power spectrum corresponding to each frequency point; Indicates the first The ratio of the average frequency of the frame signal; In step S9, a deviation compensation factor is generated for the noise estimation result of the current frame signal, and the expression for the noise estimation result of the current frame signal after compensation is obtained as follows: in Indicates the first Frame signal Noise estimation results after compensation at each frequency point; Indicates the first Frame signal Noise estimation results for each frequency point; Indicates the first Frame signal The bias compensation factor for noise estimation results at each frequency point; It represents the logarithm with the natural constant e as the base; Indicates the first Frame signal Normalized variance of the second-order statistics of the power spectrum values ​​after smoothing at each frequency point; This represents the Digamma function.

2. The fish farming cage noise tracking method based on optimal smoothing and minimum statistics according to claim 1, characterized in that, The specific method for converting the sound signal from the fish farming area to the frequency domain after framing in step S1 includes: The sound signal from the fish farming area was divided into frames of 20ms each with a frame shift of 10ms. After adding a Hamming window to each frame, a fast Fourier transform was performed to obtain the frame-level frequency domain signal.

3. The fish farming cage noise tracking method based on optimal smoothing and minimum statistics according to claim 1, characterized in that, The expression for calculating the posterior signal-to-noise ratio of the current frame signal based on the smoothed power spectrum value of the previous frame signal in step S4 is as follows: in Indicates the first Frame signal Posterior signal-to-noise ratio at each frequency point; Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal Noise power spectrum at each frequency point.

4. The fish farming cage noise tracking method based on optimal smoothing and minimum statistics according to claim 3, characterized in that, In step S4, the expression for the initial value of the smoothing factor in the smoothing process, obtained based on the posterior signal-to-noise ratio of the current frame signal, is as follows: in Indicates the first Frame signal The initial value of the smoothing factor in the smoothing process for the power spectrum corresponding to each frequency point.

5. The fish farming cage noise tracking method based on optimal smoothing and minimum statistics according to claim 4, characterized in that, In step S5, the expression for calculating the ratio of the frequency average value of the current frame signal to the sum of the smoothed power spectral values ​​of the previous frame signal and the sum of the power spectral values ​​of the current frame signal is as follows: in Indicates the first The ratio of the average frequency of the frame signal; Indicates the first The sum of the power spectral density values ​​after smoothing the frame signal; Indicates the first The sum of the power spectral density values ​​corresponding to the frame signal.

6. The fish farming cage noise tracking method based on optimal smoothing and minimum statistics according to claim 1, characterized in that, The expression for smoothing the power spectrum of the current frame signal based on the final value of the smoothing factor during the smoothing process in step S7 is as follows: in Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The final value of the smoothing factor for each frequency point during the smoothing process; Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; Show the first Frame signal The power spectrum value corresponding to each frequency point.

7. The fish farming cage noise tracking method based on optimal smoothing and minimum statistics according to claim 1, characterized in that, The expression for obtaining the minimum power spectrum value after smoothing the first L frames of signal based on the minimum statistics method in step S8 is as follows: in Indicates the first Frame signal Noise estimation results for each frequency point; Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point; This indicates taking the minimum value.

8. The fish farming cage noise tracking method based on optimal smoothing and minimum statistics according to claim 7, characterized in that, The expression for the normalized variance of the second-order statistic of the smoothed power spectrum value of the current frame signal in step S9 is as follows: in Indicates the first Frame signal Normalized variance of the second-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The second-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The first-order statistics of the power spectrum values ​​after smoothing at each frequency point; These are the weighting coefficients; Indicates the first Frame signal The second-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal The square of the power spectrum value after smoothing at each frequency point; Indicates the first Frame signal The first-order statistics of the power spectrum values ​​after smoothing at each frequency point; Indicates the first Frame signal Power spectrum values ​​after smoothing at each frequency point.

Images