A method for detecting whistle communication signals of wild Chinese white dolphins

Abstract

The present invention relates to a wild Chinese white dolphin whistle communication signal detection method, comprising: obtaining a sound signal in the ocean; framing the sound signal, and calculating the short-time energy of each frame of the sound signal; determining an energy threshold according to the fluctuation range of the short-time energy; performing a short-time Fourier transform on each frame of the sound signal to obtain a time-frequency component matrix, and extracting a portion corresponding to a target frequency from the time-frequency component matrix as a matrix to be tested; traversing the sequence of all time points in the matrix to be tested, extracting a time point sequence exceeding the energy threshold, finding a time point with energy greater than a preset component from the time point sequence, and calculating the concentration range of the high-energy frequency component at the time point, determining whether it is the same whistle communication signal according to the concentration range of the high-energy frequency wind volume; extracting a time point signal that meets the duration of the wild Chinese white dolphin whistle communication signal. The present invention can detect the sound signal of a dolphin in real time in a marine environment.

Description

Technical Field

[0001] The invention relates to the technical field of cetacean detection, and in particular to a method for detecting whistle communication signals of wild Chinese white dolphins. Background Art

[0002] The Chinese white dolphin (Sousa chinensis) is a first-class protected whale species in my country. It is mainly distributed in the shallow waters along the coast of the western Pacific and Indian Oceans. In Chinese waters, it is mainly distributed in: the coastal areas of Ningde, Xiamen, the west side of Taiwan Island, the coastal areas of Shantou, the waters of the Pearl River Estuary, Leizhou Bay in Guangdong, Beibu Gulf in Guangxi, and the waters east of Hainan. With the continuous development of my country's marine strategy, the habitat of the Chinese white dolphin is highly overlapped with the scope of human activities, which has a certain degree of impact on the survival of the Chinese white dolphin.

[0003] Similar to other marine mammals, Chinese white dolphins rely on the propagation characteristics of sound for hunting, navigation, and detection. Therefore, surveying Chinese white dolphins through acoustics is more efficient than image collection or manual positioning and tracking. The sound signals of Chinese white dolphins are highly correlated with their own existence, and passive acoustics has no impact on the marine environment and other organisms. It is a sustainable and efficient method.

[0004] Common endpoint detection methods include: double threshold method, which uses short-time energy and short-time zero-crossing rate for detection. The short-time energy calculation formula is:

[0005]

[0006] In the formula, E i is the short-time energy of the i-th frame, N is the number of sampling points of a single frame signal, x n (m) is the amplitude of each sampling point. Setting two thresholds based on short-time energy and short-time zero-crossing rate can better distinguish voiced, unvoiced and silent sounds, but the noise field and other biological sounds in the ocean environment will also cause the zero-crossing rate to change, which cannot be used as a parameter for dolphin sound detection. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a method for detecting whistle communication signals of wild Chinese white dolphins, which can detect the sound signals of dolphins in real time in a marine environment.

[0008] The technical solution adopted by the present invention to solve the technical problem is: to provide a method for detecting whistle communication signals of wild Chinese white dolphins, comprising the following steps:

[0009] Acquire sound signals in the ocean;

[0010] Divide the sound signal into frames and calculate the short-time energy of each frame of the sound signal;

[0011] Determining an energy threshold according to the fluctuation range of the short-time energy;

[0012] Performing a short-time Fourier transform on each frame of the sound signal to obtain a time-frequency component matrix, and extracting a portion corresponding to the target frequency from the time-frequency component matrix as a matrix to be tested;

[0013] Traversing the sequence of all time points in the matrix to be tested, extracting the sequence of time points that exceed the energy threshold, finding the time point whose energy is greater than the preset component from the sequence of time points, and calculating the concentration range of the high-energy frequency component at the time point, and determining whether it is the same whistle communication signal according to the concentration range of the high-energy frequency wind volume;

[0014] Extract the time point signal that meets the duration of the wild Chinese white dolphin whistle communication signal.

[0015] When the sound signal is framed, 1 / 100 to 1 / 50 of the duration of the wild Chinese white dolphin whistle communication signal is selected as the frame length.

[0016] When determining the energy threshold according to the fluctuation range of the short-time energy, the 2 / 3 digit of all the short-time energies is used as the energy threshold.

[0017] The traversing of the sequence of all time points in the matrix to be tested, extracting the sequence of time points exceeding the energy threshold, finding the time point whose energy is greater than the preset component from the sequence of time points, and calculating the concentration range of the high-energy frequency component at the time point, and determining whether it is the same whistle communication signal according to the concentration range of the high-energy frequency wind volume, includes:

[0018] If the energy at a time point does not exceed the preset component, the time point is regarded as the end of the effective sound segment;

[0019] If the energy at a time point exceeds the preset component, the frequency range of the high-energy frequency component at that time point is calculated. If the high-energy frequency component at that time point is scattered in multiple different frequencies, then that time point is regarded as the end of the valid sound segment; if the high-energy frequency component at that time point is concentrated near a certain frequency, then it is compared with the frequency range calculated at the previous time point. If the difference between the two exceeds the threshold, the previous record is valid, and the time point is recorded as the beginning of the next sound segment; if the two are continuous, then continue to record the time point and detect the next time point.

[0020] The energy at the time point exceeding the preset component means that the energy at the time point is greater than the high-energy frequency component of 90% of the components.

[0021] The time point signal that meets the duration of the wild Chinese white dolphin whistle communication signal is extracted as follows:

[0022] The duration of the effective sound segment is extracted, and the part that does not meet the duration of the wild Chinese white dolphin whistle communication signal is removed to obtain the time point signal that meets the duration of the wild Chinese white dolphin whistle communication signal.

[0023] Beneficial Effects

[0024] Due to the adoption of the above technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the present invention combines the high-energy frequency component concentration range, duration and short-time energy for aquatic organism endpoint detection, based on passive acoustic technology, and will not affect dolphins and other marine organisms; the three characteristics used in the method, short-time energy and high-energy frequency component concentration range can detect signals that meet the energy characteristics of vocal behavior in the marine environment and signals that meet the frequency range of the whistle communication (referred to as whistle) signal of the Chinese white dolphin. The combination of the two can obtain the vocal behavior that meets the sound frequency of the Chinese white dolphin, and then according to the duration characteristics of the whistle signal of the Chinese white dolphin, the duration of each extracted signal is tested. These three characteristics complement each other to improve the accuracy of detection; the method has strong adaptability in the marine environment sound field, can meet the sound endpoint detection of the Chinese white dolphin under different signal-to-noise ratio conditions, and the detection accuracy can reach more than 95%; the use of this method can protect dolphins, analyze the vocalization habits of dolphins whistles, and lay the foundation for deep learning research on the production of data sets of Chinese white dolphin communication signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of an embodiment of the present invention;

[0026] Figure 2 It is the energy distribution scatter diagram of the ocean ambient noise field;

[0027] Figure 3 It is a detection effect diagram, in which adjacent dotted lines represent the starting point and the ending point of a whistle signal;

[0028] Figure 4 It is the spectrogram of whistle detection result;

[0029] Figure 5 This is a time-frequency analysis diagram of the whistle signal of the Chinese white dolphin. DETAILED DESCRIPTION

[0030] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0031] The embodiment of the present invention relates to a method for detecting a whistle communication signal of a wild Chinese white dolphin, such as Figure 1 As shown, the following steps are included:

[0032] Step 1: Acquire sound signals in the ocean; wherein the sound signals include whistle communication signals of wild Chinese white dolphins.

[0033] Step 2, frame the sound signal and calculate the short-time energy of each frame of the sound signal; when framing in this step, select 1 / 100 to 1 / 50 of the duration of the wild Chinese white dolphin whistle communication signal as the frame length.

[0034] Step 3, determine the energy threshold according to the fluctuation range of the short-time energy; the energy threshold in this step can be obtained by observing the noise field in the marine environment. Noise is usually evenly distributed on the time axis, with large energy, wide and stable spectrum distribution range. The dolphin sound occupies a small proportion on the time axis, and the range of the noise spectrum distribution in the frequency domain is higher and more concentrated. Therefore, the energy threshold can be determined by comparing the short-time energy fluctuation range according to the noise distribution characteristics. Considering the propagation characteristics of sound, the high-frequency component decays faster, so the 2 / 3 digit of all the short-time energies is used as the energy threshold.

[0035] Step 4: Perform short-time Fourier transform on each frame of sound signal to obtain a time-frequency component matrix, and extract the part corresponding to the target frequency from the time-frequency component matrix as the matrix to be tested;

[0036] Step 5, traverse the sequence of all time points in the matrix to be tested, extract the sequence of time points that exceed the energy threshold, find the time point whose energy is greater than the preset component from the time point sequence, and calculate the concentration range of the high-energy frequency component at the time point, and determine whether it is the same whistle communication signal according to the concentration range of the high-energy frequency wind volume; specifically, if the energy at the time point does not exceed the preset component, the time point is used as the end of the effective sound segment; if the energy at the time point exceeds the preset component, the frequency range of the high-energy frequency component at the time point is calculated, and if the high-energy frequency component at the time point is scattered in multiple different frequencies, the time point is used as the end of the effective sound segment; if the high-energy frequency component at the time point is concentrated near a certain frequency, it is compared with the frequency range calculated at the previous time point. If the difference between the two exceeds the threshold, the previous record is valid, and the time point is recorded as the beginning of the next sound segment; if the two are continuous, continue to record the time point and detect the next time point. Among them, the energy at the time point exceeds the preset component means that the energy at the time point is greater than the high-energy frequency component of 90% of the components.

[0037] Step 6, extract the time point signal that meets the duration of the whistle communication signal of the wild Chinese white dolphin. In this step, the duration of the effective sound segment is extracted, and the part that does not meet the duration of the whistle communication signal of the wild Chinese white dolphin is removed to obtain the time point signal that meets the duration of the whistle communication signal of the wild Chinese white dolphin.

[0038] The present invention is further illustrated by a specific example below.

[0039] 1. Get the data

[0040] The experimental data comes from the sound signals of wild Chinese white dolphins collected from Sanniang Bay, Guangxi. The experimental platform of this embodiment is python and Audition.

[0041] 2. Data analysis and processing

[0042] The data sampling frequency is 288KHz. In order to facilitate the observation of whistle signals, the signals are framed. Framing is essentially windowing in the time domain, and then the windowed signals are analyzed in the frequency domain. The longer the frame length, the higher the frequency resolution and the lower the time resolution. Since the duration of the whistle signal of the Chinese white dolphin is 0.3 to 0.9s and the frequency is concentrated in 5000 to 30000Hz, the requirements for frequency resolution to extract dolphin signals are higher than the requirements for time resolution. Therefore, the frame length should be selected as a smaller value as possible. 0.03s is selected as the frame length for framing, and short-time Fourier transform is performed to obtain a matrix of time-frequency components. The part corresponding to the frequency of 5000 to 30000Hz in the matrix is ​​extracted as the matrix to be tested.

[0043] 3. Determine the threshold

[0044] By observing the distribution characteristics of noise on the time axis (see Figure 2 ), the threshold is determined by comparing the short-time energy fluctuation range. Considering the propagation characteristics of sound, the high-frequency component decays faster, and the short-time energy threshold is proposed to be 2 / 3 of the entire signal energy.

[0045] 4. Conduct testing

[0046] Traverse the sequence of all time points in the matrix to be tested, extract the time point sequence that exceeds the short-time energy threshold, find the high-energy frequency component whose energy is greater than 90% of the components at this time point, and calculate the concentration range of its high-energy frequency component. When the concentration range of high-energy frequency components between adjacent time points differs within 1000Hz, it is considered to be the same whistle signal. When the concentration range of high-energy frequency components between adjacent time points differs by more than 1000Hz, they are considered to be different whistle signals. Figure 3 This is the detection effect diagram.

[0047] 5. Verify the test results

[0048] All detected sound segments are tested, and the duration of all extracted sound segments is calculated by recording the time points, and sound segments with a duration of less than 0.3s are eliminated. Figure 4 It is the spectrogram of whistle detection results.

[0049] 6. Generate valid sound segments

[0050] The collected signal is cut according to the time point of the detection result, and the cut file is saved as the whistle signal sound of the Chinese white dolphin. The collected original signal is deleted from the cache file list to speed up the next read of the cache folder. After the processing is completed, the statistical results are shown in Table 1. Figure 5 This is a time-frequency analysis diagram of the whistle signal of the Chinese white dolphin.

[0051] Table 1 Test results

[0052] Number of tests The number of detected errors total Whistle signal 135 0 135

[0053] 7. Comparison with other methods

[0054] The double threshold method, the endpoint detection method proposed by Harma and the method proposed in this embodiment are used to detect the same section of Chinese white dolphin whistle signal, and the results are compared as shown in Table 2.

[0055] Table 2 Comparison of detection results of the same signal using different methods

[0056] method Detected whistle signal The average duration of noise included Lost whistle signal Double Threshold Method 4 0.37s 2 Harma detection method 2 0.25s 4 This embodiment 6 0.14s 0

[0057] It is not difficult to find that the detection method of the present invention will not affect dolphins, and can perform endpoint detection of the whistle signal of the Chinese white dolphin in the marine environmental noise field; the present invention can avoid the interference of noise on the detection, has a fast speed, a small error and does not require human intervention, thereby improving the efficiency of the whistle signal extraction of the Chinese white dolphin.

Claims

1. A method for detecting whistle communication signals of wild Chinese white dolphins, characterized in that: The following steps are involved: Acquire sound signals in the ocean; Divide the sound signal into frames and calculate the short-time energy of each frame of the sound signal; Determining an energy threshold according to the fluctuation range of the short-time energy; Performing a short-time Fourier transform on each frame of the sound signal to obtain a time-frequency component matrix, and extracting a portion corresponding to the target frequency from the time-frequency component matrix as a matrix to be tested; Traversing the sequence of all time points in the matrix to be tested, extracting the sequence of time points that exceed the energy threshold, finding the time point whose energy is greater than the preset component from the sequence of time points, and calculating the concentration range of the high-energy frequency component at the time point, and determining whether it is the same whistle communication signal according to the concentration range of the high-energy frequency component, specifically including: If the energy at a time point does not exceed the preset component, the time point is regarded as the end of the effective sound segment; If the energy at a time point exceeds the preset component, the frequency range of the high-energy frequency component at that time point is calculated. If the high-energy frequency component at that time point is scattered in multiple different frequencies, that time point is regarded as the end of the valid sound segment; if the high-energy frequency component at that time point is concentrated near a certain frequency, it is compared with the frequency range calculated at the previous time point. If the difference between the two exceeds the threshold, the previous record is valid, and the time point is recorded as the beginning of the next sound segment; if the two are continuous, the time point is continued to be recorded, and the next time point is detected; Extract the time point signal that meets the duration of the wild Chinese white dolphin whistle communication signal.

2. The method for detecting whistle communication signals of wild Chinese white dolphins according to claim 1, characterized in that: When the sound signal is framed, 1 / 100 to 1 / 50 of the duration of the wild Chinese white dolphin whistle communication signal is selected as the frame length.

3. The method for detecting whistle communication signals of wild Chinese white dolphins according to claim 1, characterized in that: When determining the energy threshold according to the fluctuation range of the short-time energy, the 2 / 3 digit of all the short-time energies is used as the energy threshold.

4. The method for detecting whistle communication signals of wild Chinese white dolphins according to claim 1, characterized in that: The energy at the time point exceeding the preset component means that the energy at the time point is greater than the high-energy frequency component of 90% of the components.

5. The method for detecting whistle communication signals of wild Chinese white dolphins according to claim 1, characterized in that: The time point signal that meets the duration of the wild Chinese white dolphin whistle communication signal is extracted as follows: The duration of the effective sound segment is extracted, and the part that does not meet the duration of the wild Chinese white dolphin whistle communication signal is removed to obtain the time point signal that meets the duration of the wild Chinese white dolphin whistle communication signal.

Images