Automatic whale guinea pig whistle recognition method, program, equipment and storage medium

Abstract

The invention belongs to the technical field of marine organism sound detection, and particularly relates to a whale dolphin whistle automatic identification method, a program, equipment and a storage medium. According to the invention, noise signals are eliminated by using a sum-based time domain envelope detection method, so that the influence of impulse noise on detection is prevented; detecting end points of the signals by adopting a double-threshold end point detection algorithm, and extracting signal segments; the CW or LFM signal is eliminated by using the characteristic that the CW signal power spectrum has only one main spectral line and the characteristic that the sliding window instantaneous frequency of the LFM signal is integrally kept in a monotone increasing state along with time, and finally, whether the whistling of the whale dolphin is judged through surface GM-PHD spectrum contour extraction. According to the method, the workload of eliminating non-whale-guinea-pig whistle interference through manual sound listening can be reduced, and the accuracy and the working efficiency of whale-guinea-pig whistle recognition are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine biological sound detection, and in particular relates to a method, program, device and storage medium for automatic recognition of cetacean whistles. Background Art

[0002] The ocean is home to many organisms that use sound to communicate, locate prey, avoid predators, or perform other behaviors. The characteristics of these sounds vary depending on the species, behavior, environment, and seasonal changes. Marine life calls are of great significance in the marine environment. They can reflect the distribution, activity patterns, and ecological status of biological populations. Therefore, by analyzing these sounds, information about biodiversity and the health of marine ecosystems can be obtained. In recent years, with the advancement of hydroacoustic technology and the improvement of computing power, the technology for detecting marine life calls has developed significantly. From simple hydroacoustic recording to complex automatic sound recognition systems, scientists are able to capture, record, and analyze marine life sounds more accurately. However, in the detection and identification of animal whistle signals such as cetaceans, the identification of cetacean calls is often interfered by the hydroacoustic signals emitted by other platforms and the interference of marine environmental noise. Therefore, identifying cetacean whistles from various types of detected hydroacoustic signals and marine background noise is conducive to the statistical analysis of the characteristics of cetaceans.

[0003] Chinese patent CN117275491B discloses a call classification method based on audio conversion and temporal attention graph neural network. This method mainly proposes a neural network classification method for cetacean calls to identify cetacean calls. At present, the patent technology publicly released in China lacks the recognition and detection of cetacean calls under the interference of underwater acoustic communication signals. Therefore, it is of great research value to design a set of automatic recognition methods for cetacean whistles and complete the statistical analysis of the characteristics of cetaceans. Summary of the invention

[0004] The purpose of the present invention is to provide a method, program, device and storage medium for automatically identifying cetacean whistles, which can identify and detect cetacean whistles under underwater acoustic communication signals and marine environment background noise.

[0005] A method for automatically identifying cetacean whistles comprises the following steps:

[0006] Step 1: Obtain the signal to be detected, and use the time domain envelope detection method based on the sum to eliminate the noise signal; detect the endpoints of the signal and extract the signal fragments;

[0007] Step 2: Calculate the power spectrum, and use the feature that the CW signal power spectrum has only one main spectrum line to identify and exclude the CW signal fragment;

[0008] Step 3: Calculate the instantaneous frequency increment ratio, identify and exclude LFM signal segments;

[0009] Step 4: Extract multiple groups of instantaneous frequencies of the signal segment based on the GM-PHD method, use the polynomial fitting method to obtain the fitting value of the instantaneous frequency, calculate the square mean of the distance between each group of instantaneous frequencies and the fitting value, and select the largest square mean as the eigenvalue; if the eigenvalue is less than the detection threshold, the signal segment is identified as a cetacean whistle.

[0010] Furthermore, in step 1, the noise signal is eliminated by using a time domain envelope detection method based on the sum, specifically:

[0011] Add a window to the detected signal and perform short-time FFT. Take the sum of the maximum values ​​of the short-time FFT of n sliding windows S. sum As the test statistic, the calculation formula is:

[0012]

[0013] Among them, n is the total number, S m (t) is the short-time FFT maximum value corresponding to time t, and N is the number of summations;

[0014] The detection statistic is compared with a preset signal detection threshold. If the detection statistic is less than the threshold, it is determined to be a noise signal.

[0015] Furthermore, in step 1, a double threshold endpoint detection algorithm is used to detect the endpoints of the signal to obtain the sample point positions in the signal segment that are greater than the first threshold; for each sample point position that is greater than the threshold, the position is used as the endpoint to continue downward to find a position that is less than the second threshold as another endpoint, the time difference between the two endpoints is calculated as the detected signal duration, and the signal segment between the two endpoints is extracted.

[0016] Furthermore, the calculation method of the instantaneous frequency increment ratio in step 3 is:

[0017] Perform short-time Fourier transform on the signal segment and calculate the instantaneous frequency of the sliding window. Differencing the M instantaneous frequencies obtained obtains M-1 frequency differences. The instantaneous frequency increment ratio is f. m The calculation formula is:

[0018]

[0019] Among them, f d Represents the instantaneous frequency corresponding to the dth sliding window; the sliding window instantaneous frequency of the LFM signal remains monotonically increasing over time. If it is an LFM signal, the N-1 instantaneous frequency self-increase ratios of the signal segment are monotonically changing; otherwise, it is not an LFM signal.

[0020] Furthermore, the step 4 is specifically as follows:

[0021] The instantaneous frequency f at a certain moment is regarded as the target to be tracked, and the GM-PHD filter is used to track the spectrum profile. The state vector x of the spectrum profile is k It is composed of the instantaneous frequency f and the rate of change of frequency over time α, that is, x k =[f,α] T , then the state model is:

[0022]

[0023] Where Δ represents the step size of the short-time Fourier transform; v k ~N(0,Σ v ) means the mean is zero and the covariance is Σ v Gaussian noise; in the process of extracting the spectrum profile of the whistle signal of cetaceans, only the frequency quantity is useful information, so the observation model is:

[0024] f k =Hx k +v k

[0025] Wherein, the measurement matrix H = [1 0];

[0026] For each extracted instantaneous frequency f of the signal segment k Using the polynomial fitting method, the fitting value is f k ′, and then calculate the instantaneous frequency f k and the fitted value f k ′The square mean of the distance, select the largest square mean as the eigenvalue;

[0027]

[0028] Where Q is the instantaneous frequency f in the spectrum profile extracted from the signal segment k The total number of .

[0029] A computer device / equipment / system comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-mentioned method for automatically identifying cetacean whistles.

[0030] A computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the above-mentioned method for automatically identifying cetacean whistles.

[0031] A computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned method for automatically identifying cetacean whistles.

[0032] The beneficial effects of the present invention are:

[0033] The present invention provides a complete set of cetacean whistle interference identification process for received signals, adopts the time domain envelope detection method based on the sum to exclude noise signals, and prevents the influence of pulse noise on the detection; adopts the double threshold endpoint detection algorithm to detect the endpoints of the signal and extract the signal fragments; utilizes the characteristics that the power spectrum of the CW signal has only one main spectrum line, and the sliding window instantaneous frequency of the LFM signal maintains a monotonically increasing state over time as a whole to exclude CW or LFM signals, and finally extracts the GM-PHD spectrum profile to determine whether it is a cetacean whistle. The present invention can reduce the workload of artificial listening to exclude the interference of non-cetacean whistles, and improve the accuracy and work efficiency of cetacean whistle identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The present invention is a flow chart of a method for automatically identifying cetacean whistles.

[0035] Figure 2 It is a time-frequency diagram of the signal set to be detected in an embodiment of the present invention.

[0036] Figure 3 These are four whistle spectrum contour diagrams extracted using the GM-PHD method in an embodiment of the present invention.

[0037] Figure 4 This is a recognition result diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The present invention is further described below in conjunction with the accompanying drawings.

[0039] The present invention identifies and detects cetacean whistles under underwater acoustic communication signals and ocean environment background noise, and provides a method, program, device and storage medium for automatic identification of cetacean whistles, which is beneficial to improving the accuracy and work efficiency of cetacean whistle identification and realizing a complete and automatic identification process of cetacean call signals.

[0040] A method for automatically identifying cetacean whistles comprises the following steps:

[0041] Step 1: Obtain the signal to be detected, and use the time domain envelope detection method based on the sum to eliminate the noise signal; detect the endpoints of the signal and extract the signal fragments;

[0042] Add a window to the detected signal and perform short-time FFT. Take the sum of the maximum values ​​of the short-time FFT of n sliding windows S. sum As the test statistic, the calculation formula is:

[0043]

[0044] Among them, n is the total number, S m (t) is the short-time FFT maximum value corresponding to time t, and N is the number of summations;

[0045] Compare the detection statistic with a preset signal detection threshold, and if the detection statistic is less than the threshold, determine it as a noise signal;

[0046] The double threshold endpoint detection algorithm is used to detect the endpoints of the signal and obtain the sample point positions in the signal segment that are greater than the first threshold. For each sample point position greater than the threshold, the position is used as the endpoint to continue to search downward to a position less than the second threshold as the other endpoint. The time difference between the two endpoints is calculated as the detected signal duration, and the signal segment between the two endpoints is extracted.

[0047] Step 2: Calculate the power spectrum, and use the feature that the CW signal power spectrum has only one main spectrum line to identify and exclude the CW signal fragment;

[0048] Step 3: Calculate the instantaneous frequency increment ratio, identify and exclude LFM signal segments;

[0049] Perform short-time Fourier transform on the signal segment and calculate the instantaneous frequency of the sliding window. Differencing the M instantaneous frequencies obtained obtains M-1 frequency differences. The instantaneous frequency increment ratio is f. m The calculation formula is:

[0050]

[0051] Among them, f d Indicates the instantaneous frequency corresponding to the dth sliding window; the sliding window instantaneous frequency of the LFM signal remains monotonically increasing over time. If it is an LFM signal, the N-1 instantaneous frequency self-increase ratios of the signal segment are monotonically changing; otherwise, it is not an LFM signal;

[0052] Step 4: Based on the GM-PHD (Gaussian mixture probability hypothesis density filter) method, multiple groups of instantaneous frequencies of the signal segment are extracted, and the fitting value of the instantaneous frequency is obtained by the polynomial fitting method. The square mean of the distance between each group of instantaneous frequencies and the fitting value is calculated, and the largest square mean is selected as the eigenvalue; if the eigenvalue is less than the detection threshold, the signal segment is identified as a cetacean whistle.

[0053] The instantaneous frequency f at a certain moment is regarded as the target to be tracked, and the GM-PHD filter is used to track the spectrum profile. The state vector x of the spectrum profile is k It is composed of the instantaneous frequency f and the rate of change of frequency over time α, that is, x k =[f,α] T , then the state model is:

[0054]

[0055] Where Δ represents the step size of the short-time Fourier transform; v k ~N(0,Σ v ) means the mean is zero and the covariance is Σ v Gaussian noise; in the process of extracting the spectrum profile of the whistle signal of cetaceans, only the frequency quantity is useful information, so the observation model is:

[0056] f k =Hx k +v k

[0057] Wherein, the measurement matrix H = [1 0];

[0058] For each extracted instantaneous frequency f of the signal segment k Using the polynomial fitting method, the fitting value is f k ′, and then calculate the instantaneous frequency f k and the fitted value f k ′The square mean of the distance, select the largest square mean as the eigenvalue;

[0059]

[0060] Where Q is the instantaneous frequency f in the spectrum profile extracted from the signal segment k The total number of .

[0061] Embodiment 1:

[0062] The following is an example of a signal composed of PSK, FSK, OFDM, DSSS, ship radiation noise, broadband signal, CW, LFM and four segments of cetacean calls.

[0063] (1) Sampling frequency 48000 Hz, communication frequency band 6-10 kHz, CW signal frequency 8 kHz, LFM frequency modulation slope 6 kHz, ship noise duration 4 s, four different cetacean sound clips.

[0064] (2) Detect whether there is a signal. The signal detection statistic is determined by using the sum-based time-domain envelope detection method and compared with the threshold. The sum detection method improves the classic short-time Fourier transform detection method. The specific improvement idea is: add a window to the received signal and perform a short-time FFT, and take the sum of the maximum values ​​of the n sliding window short-time FFTs S sumAs new detection statistics, 9 different detection statistics are obtained as follows: 148, 75, 178, 182, 209, 77, 77, 74, 73. The double threshold endpoint detection algorithm is used to detect the endpoints of the signal, and the position of the i-th sample point in the signal segment that is greater than the threshold can be obtained. From this position as the starting point, continue to search downward to the position less than the threshold, and calculate the time difference between the two endpoints as the detected signal duration t i .

[0065] (3) CW signal identification: Calculate the power spectrum and use the fact that the CW signal power spectrum has only one main spectrum line to identify the CW signal.

[0066] (4) LFM signal identification. The concept of instantaneous frequency increment ratio is used to identify linear frequency modulation signals. If it is an LFM signal, the N-1 instantaneous frequency increment ratios of this signal should change monotonically, otherwise it is not an LFM signal.

[0067] (5) Extract the instantaneous frequency of the signal based on the GM-PHD (Gaussian mixture probability hypothesis density filter) method. For each extracted instantaneous frequency, a polynomial fitting method is used, where the number of polynomial fittings is 7, and then the square mean of the distance between the instantaneous frequency and the fitting value is calculated, and the largest square mean is selected as the eigenvalue. The eigenvalue of each segment is obtained as follows Figure 4 As shown in the figure, the eigenvalues ​​of the first four segments are all greater than 0.001, that is, greater than the threshold Th, so they are judged as underwater acoustic communication signals, and the starting time and duration are marked for subsequent underwater acoustic signal identification; the middle four segments do not give eigenvalues ​​because they are noise or CW, LFM and other signals, which are excluded in advance, and the spectrum profile is not extracted and the eigenvalue is not calculated; the eigenvalues ​​of the last four segments are all less than 0.001, so they are judged as biological call signals and marked. The above results are consistent with the generated signal type, which can basically determine the correctness of the automatic recognition process of cetacean whistles.

[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for automatically identifying cetacean whistles, characterized in that: The following steps are involved: Step 1: Obtain the signal to be detected and use the time domain envelope detection method based on the sum to eliminate the noise signal; Detect the endpoints of the signal and extract the signal fragments; Step 2: Calculate the power spectrum, and use the feature that the CW signal power spectrum has only one main spectrum line to identify and exclude the CW signal fragment; Step 3: Calculate the instantaneous frequency increment ratio, identify and exclude LFM signal segments; Step 4: Extract multiple groups of instantaneous frequencies of signal segments based on the GM-PHD method, obtain the fitting value of the instantaneous frequency by using the polynomial fitting method, calculate the square mean of the distance between each group of instantaneous frequencies and the fitting value, and select the largest square mean as the eigenvalue; If the eigenvalue is less than the detection threshold, the signal segment is identified as a cetacean whistle.

2. The method for automatically identifying cetacean whistles according to claim 1, characterized in that: In step 1, the noise signal is eliminated by using a time domain envelope detection method based on the sum, specifically: Add a window to the detected signal and perform short-time FFT, and take the sum of the maximum values ​​of the short-time FFT of n sliding windows S sum As the test statistic, the calculation formula is: Among them, n is the total number, S m (t) is the short-time FFT maximum value corresponding to time t, and N is the number of summations; The detection statistic is compared with a preset signal detection threshold. If the detection statistic is less than the threshold, it is determined to be a noise signal.

3. The method for automatically identifying cetacean whistles according to claim 1, characterized in that: In the step 1, a double threshold endpoint detection algorithm is used to detect the endpoints of the signal to obtain the sample point positions in the signal segment that are greater than the first threshold. For each sample point position that is greater than the threshold, the position is used as the endpoint to continue downward to find a position that is less than the second threshold as another endpoint, and the time difference between the two endpoints is calculated as the detected signal duration, and the signal segment between the two endpoints is extracted.

4. The method for automatically identifying cetacean whistles according to claim 1, characterized in that: The calculation method of the instantaneous frequency increment ratio in step 3 is: Perform short-time Fourier transform on the signal segment and calculate the instantaneous frequency of the sliding window. Differencing the M instantaneous frequencies obtained obtains M-1 frequency differences. The instantaneous frequency increment ratio is f. m The calculation formula is: Among them, f d Represents the instantaneous frequency corresponding to the dth sliding window; the sliding window instantaneous frequency of the LFM signal remains monotonically increasing over time. If it is an LFM signal, the N-1 instantaneous frequency self-increase ratios of the signal segment are monotonically changing; otherwise, it is not an LFM signal.

5. The method for automatically identifying cetacean whistles according to claim 1, characterized in that: The step 4 is specifically as follows: The instantaneous frequency f at a certain moment is regarded as the target to be tracked, and the GM-PHD filter is used to track the spectrum profile. The state vector x of the spectrum profile is k It is composed of the instantaneous frequency f and the rate of change of frequency over time α, that is, x k =[f,α] T , then the state model is: Where Δ represents the step size of the short-time Fourier transform; v k ~N(0,Σ v ) means the mean is zero and the covariance is Σ v Gaussian noise; in the process of extracting the spectrum profile of the whistle signal of cetaceans, only the frequency quantity is useful information, so the observation model is: f k =Hx k +v k Wherein, the measurement matrix H = [1 0]; For each extracted instantaneous frequency f of the signal segment k Using the polynomial fitting method, the fitting value is f k ′, and then calculate the instantaneous frequency f k and the fitted value f k ′The square mean of the distance, select the largest square mean as the eigenvalue; Where Q is the instantaneous frequency f in the spectrum profile extracted from the signal segment k The total number of .

6. A computer device / equipment / system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images