CN115150009B - A dolphin echolocation signal detection method for PAM system - Google Patents

Abstract

The present invention relates to a dolphin echolocation signal detection method applied to a PAM system, belonging to the field of acoustic signals, and the steps of the method include collecting sound data, bandpass filtering the collected sound data, setting a dynamic signal detection threshold value TH, grouping data with signal amplitudes higher than the signal detection threshold value, each group of data forming a complete signal, and determining the start and end positions of each suspected signal. The method does not require FFT transformation and frequency domain analysis on the premise of ensuring the accuracy, thereby greatly improving the calculation speed and effectively reducing the system computing power consumption, thereby extending the working time of the PAM system; the algorithm combines the time domain characteristics and biological activity characteristics of the dolphin echolocation signal, thereby greatly improving the detection accuracy and work efficiency.

Description

A dolphin echolocation signal detection method for PAM system

Technical Field

The invention belongs to the field of acoustic signals, and in particular relates to a dolphin echolocation signal detection method applied to a PAM system.

Background Art

Passive Acoustic Monitoring (PAM) systems are used to monitor marine mammals in the ocean and have been increasingly used to monitor dolphin acoustic signals. Most PAM systems detect the echolocation clicks that dolphins use to navigate and detect prey, and some can detect low-frequency communication sounds such as whistles made by dolphins. However, not all dolphins make communication sounds or whistles, such as beaked whales and porpoises. Therefore, in order to cover as many species as possible, click detectors should be used as much as possible. PAM systems need to meet the requirements of low power consumption and long-term operation, and have problems such as large storage space, slow calculation speed, and low detection accuracy.

The ocean background noise contains a large number of transient noise signals similar to dolphin sound signals, which brings many difficulties to signal detection. Existing detection technology is difficult to solve the problem of interference signal and signal separation in the time domain, so it is mainly distinguished and identified in the frequency domain, but frequency domain analysis is a computationally intensive process, especially the click signal of some species requires a data sampling rate of 400kHz or higher. The huge amount of calculation will cause the PAM system to frequently replace batteries and storage devices, and real-time monitoring is even more difficult to achieve. Obviously, reducing the computing requirements of signal detection is of great significance to reducing system power consumption and storage pressure. On the other hand, some PAM systems also have subsequent detection functions. For example, in offshore engineering waters, after detecting dolphin signals, the drive will be turned on to drive away dolphins. These application scenarios have higher requirements for detection accuracy, because once the detection is wrong, it will cause adverse effects on the environment and a large waste of system resources.

Summary of the invention

The technical problem to be solved by the present invention is to provide a dolphin echolocation signal detection method applied to the PAM system. Under the premise of ensuring the accuracy, the method does not need to perform FFT transformation and frequency domain analysis, thereby greatly improving the calculation speed and effectively reducing the system computing power consumption, thereby extending the working time of the PAM system; the algorithm combines the time domain characteristics and biological activity characteristics of the dolphin echolocation signal, thereby greatly improving the detection accuracy and work efficiency.

The present invention is achieved through the following technical solutions

A dolphin echolocation signal detection method applied to a PAM system, the detection method comprising the following three steps:

1) Collect sound data, perform bandpass filtering on the collected sound data, and then calculate the signal-to-noise ratio (SNR) using the following formula:

Among them, s represents the time series containing signal data, sqrt(mean(s ² )) is the root mean square value (RMS) of the signal series, representing the noise level, and abs(s) represents the absolute value of the signal amplitude contained in the time series;

2) Setting the threshold value TH for signal-to-noise ratio detection, where the signal-to-noise ratio is the value of the data obtained by the formula in 1); when the signal-to-noise ratio satisfies

SNR＞TH

Indicates that suspected signal data is detected. Each signal pulse on the waveform consists of a group of signal data and can be regarded as a suspected signal. On the contrary, all signals less than TH are regarded as noise. According to the number of suspected signals detected, the dynamic threshold value TH' is set and the above formula is used again to extract the signal. The calculation formula of the dynamic threshold value is:

Where s' is equal to the average noise level mean(SNR), n is the number of suspected signal data, and t is the total time length of the calculation;

3) Group the suspected signal data that are higher than the dynamic threshold value, each group of suspected signal data constitutes a complete suspected signal, and determine the start and end positions of each suspected signal. The specific method is as follows:

Set the detection window length dt, where the value of dt is the minimum interval time between two suspected signals; store all suspected signal data positions that satisfy SNR>TH into array X, taking the first suspected signal data time in array X as the starting point, record the time as t(1), and the second suspected signal data time as t(2). Determine the relationship between t(2)-t(1) and dt. If t(2)-t(1)<dt, continue to move to t(3) until t(n)-t(1)>dt. Record t(1) to t(n-1) as the time array of the first signal.

Further, one or more of the following three methods are selected to screen the suspected signals whose start and end positions are determined in step 3):

(1) Control signal duration

a. After the signals are grouped, the time information of each suspected signal is extracted, including: start time, end time, peak time, and the time length of each suspected signal is calculated and saved as a signal group to be processed;

b. Set the time length threshold as the judgment condition. The value of the time length threshold is obtained by multiplying the average time length of all suspected signals extracted by a coefficient, wherein the coefficient is 1.2-1.5;

c. Traverse all suspected signals, remove suspected signals that do not meet the time length threshold, and save the remaining suspected signals after removal for the next step of screening;

(2) Introduce the minimum time interval threshold parameter L and satisfy ΔT<L<T;

Among them, △T is the arrival time difference between some sound lines with large grazing angles and the direct signal after being reflected by the sea surface and the seabed and then reaching the receiver, and T is the time interval parameter of the dolphin echolocation signal obtained from empirical data;

Traverse all suspected signals, and from the second suspected signal onwards, determine the time difference between the next suspected signal and the previous suspected signal. If the time difference does not satisfy the formula, remove the signal until the last suspected signal, and save the remaining suspected signals after removal;

(3) Determine isolated signals

Isolated signals are eliminated from all suspected signals. The isolated signals are signals for which there is no suspected signal data within 1 second before or after the signals, or suspected signals exist but the suspected signals do not meet the characteristics of dolphin echolocation signals.

The beneficial effects of the present invention compared with the prior art are as follows:

The method is characterized in that it only relies on the time domain characteristics of the dolphin echolocation signal, does not require FFT transformation and frequency domain calculation, and has a fast overall calculation speed; since the algorithm comprehensively considers the characteristics of the dolphin echolocation signal and common interference noise, and adopts corresponding methods for processing without increasing the calculation load as much as possible, the detection accuracy is greatly improved. After testing, the accuracy of dolphin sound data collected in the wild environment can reach more than 90%; the method of the present invention requires low computing power load, greatly reduces the power consumption of the system, and improves work efficiency; it can be applied to long-term passive acoustic systems and real-time monitoring systems.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is a signal processing flow chart: (a) is a time domain waveform after bandpass filtering. After filtering, the low-frequency interference signal is greatly suppressed; (b) is a time domain waveform of a single signal. The waveform of the click signal in the figure is complete, and it can be seen that filtering has little effect on the target signal; (c) and (d) are data processed by the signal-to-noise ratio method. Each signal consists of several data points; (e) is the processing result after algorithm grouping;

Fig. 2 is a schematic diagram of signal extraction;

Figure 3 is a comparison of sound signals processed using two different methods.

DETAILED DESCRIPTION

The technical solution of the present invention is further explained below by way of embodiments, but the protection scope of the present invention is not limited in any form by the embodiments.

Example 1

1. To obtain the signal-to-noise ratio of the signal and noise, the collected sound data must first be bandpass filtered, and then the signal-to-noise ratio (SNR) is calculated using the following formula:

Where s represents the time series containing signal data, and the calculation result below the semicolon is the root mean square (RMS) of the signal series, which represents the noise level. abs(s) is the absolute value of the signal amplitude, which is divided by the noise level to get the signal-to-noise ratio.

The signal processing flow is shown in Figure 1, which shows the processing process and results of the entire data and a single signal. Figure 1(a) is a time domain waveform after bandpass filtering. After filtering, the low-frequency interference signal is greatly suppressed; Figure 1(b) is a time domain waveform of a single signal. The waveform of the click signal in the figure is complete, and it can be seen that filtering has little effect on the target signal; Figure 1(c) and Figure 1(d) are data processed by the signal-to-noise ratio method. Each signal consists of several data points; Figure 1(e) is the processing result after algorithm grouping, and 6 valid signals are detected; the specific time position of each signal is determined by its peak position, as shown in Figure 1(f).

2. Set the signal amplitude detection threshold TH. When the signal data meets

SNR＞TH

Indicates that suspected signal data is detected; on the contrary, all signals less than TH are regarded as noise. The principle of signal-to-noise ratio judgment is to process the signal amplitude, amplify the gap between the effective signal and the noise, and then obtain data greater than the threshold value, as shown in Figure 2.

The choice of threshold value will directly affect the number of detection signals. If it is too high, a large number of long-distance and non-spindle signals will be lost, and if it is too low, the false alarm rate will be greatly increased. Therefore, converting the static threshold value in the prior art into a dynamic threshold value and setting the threshold value according to the signal-to-noise ratio level of different environmental background noises can effectively reduce the false alarm rate.

The calculation method of the dynamic threshold value in this embodiment takes into account the average noise level (s'=mean(SNR)), the number of suspected signals (n) and the total calculation time length (t), which can be referred to the following formula:

3. Group the suspected signal data whose signal amplitude is higher than the dynamic threshold value. Each group of suspected signal data constitutes a complete suspected signal, and determine the start and end positions of each suspected signal. The method for determining the start and end positions is to set the detection window length dt, and the value of dt is the minimum interval time between two suspected signals. Store all data positions that satisfy SNR>TH into array X, take the time of the first suspected signal data in the array as the starting point, record the time as t(1), and record the time of the second suspected signal data as t(2), and determine the relationship between t(2)-t(1) and dt. If t(2)-t(1)<dt, continue to move to t(3) until t(n)-t(1)>dt, and record t(1) to t(n-1) as the time array of the first suspected signal, as shown in Figure 2.

4. Due to the noise interference of the collection environment (especially the wild environment), the suspected signals processed in step 3 contain a large number of transient interference signals with similar shapes and characteristics to dolphin echolocation signals. In order to improve the detection accuracy, based on the above calculation results, the following three methods are used for screening (can be used simultaneously or separately):

(1) Control signal duration

The time length of dolphin echolocation signals is a very important signal feature. According to existing data statistics, its average effective length is between 0.3 and 0.5 ms. Therefore, setting the signal time length interval as a filtering condition is a very effective means. In actual detection, it is found that some transient noise signals often show that the time length is too long or very short. Therefore, by setting the signal length threshold for screening, the probability that the interference signal is just in the time length range is low, which can greatly reduce the false alarm rate.

Specific steps:

a. After the signal data is grouped, the time information of each suspected signal is extracted, including: start time, end time, peak time, and the time length of each suspected signal is calculated and saved as a signal group to be processed;

b. Set the time length threshold as the judgment condition. The value of the time threshold is obtained by multiplying the average time length of all suspected signals extracted by the coefficient. Since the signal emitted by the dolphin is not fixed and the time length will change, in order to detect as many signals as possible, the coefficient can be 1.2-1.5.

c. Traverse all suspected signals, eliminate signals that do not meet the time length threshold, and save the remaining signals for the next step of screening.

(2) Introducing the minimum time interval threshold L as the judgment condition

After analyzing the extraction results, it was found that a large part of the false detection signals came from the sea surface/seabed reflection signals of the original signal. This type of interference signal is basically the same as the original signal in terms of various signal characteristics, so it is difficult to remove. By introducing the minimum time interval threshold parameter and setting the minimum time interval threshold between signals, the false detection of this type of interference can be effectively suppressed. Since this type of interference is generated with the real signal, it has little effect on the system detection accuracy. Specific steps:

a. The sea surface/bottom-sea reflection signal is due to the fact that some sound rays with large grazing angles are reflected by the sea surface and the bottom of the sea before reaching the receiver. There is an arrival time difference △T (determined by the local water depth) between them and the direct signal. The value of the minimum time interval threshold (L) is adjusted according to the local water depth and the acoustic characteristics of the target species. It needs to be greater than △T and less than the time interval T of the dolphin echolocation signal (the average value of the statistical data of the dolphin species is taken, and the recommended value is 0.05s). Please refer to the following formula;

ΔT<L<T

b. Traverse all suspected signals, starting from the second suspected signal, determine the time difference between the next suspected signal and the previous suspected signal. If it does not meet the above formula, then eliminate the signal until the last signal, and save the remaining suspected signals.

(3) Isolated signal judgment

Dolphin click signals often appear in clusters, so isolated signals (meaning that no other signals are detected within a certain time range before and after the signal) are most likely various interference signals. Therefore, the method of removing isolated signals can effectively remove some interference signals.

The judgment basis of isolated signal is mainly:

(1) There is no suspected signal within 1 second before or after the signal;

(2) If there is a suspected signal, but it does not meet the characteristics of a dolphin echolocation signal, it is an isolated signal.

The effect comparison of the signal processing by the above method is shown in Figure 3. Among them, a and c are dolphin signals extracted by the method of steps 1-3; b and d are the signals extracted by steps 1-3, and the seabed reflection signals of the dolphin sound signals are further removed by using the screening methods (1)-(3), wherein the interference signals of the same amplitude are removed.

Example 2

The method of Example 1 was applied to test experimental data in different environments, with 3 sets of data for each environment, and the time required for FFT transformation processing was counted under the same hardware conditions (only transformation was performed without subsequent extraction processing). The result was more than 20 times the calculation time of the algorithm. The results are shown in Table 1:

Table 1 shows the calculation time and filtering time for processing experimental data in different environments and the processing time of only FFT transformation

Note: The calculation processing time will vary according to the hardware conditions of the computer system. The data in this embodiment is processed by the same computer system.

Field environment: data obtained from sea measurement experiments conducted during dolphin activity. The experimental platform is usually a small fishing boat, which is often accompanied by a large amount of random background noise;

Indoor environment: The data is measured in indoor breeding places such as oceanariums. The experimental platform is on land and the background noise is very small.

Calculation time and filtering time: From the analysis of the results, the filtering time in the table represents the time required for the computer's digital filter. The filtering time accounts for about 80% of the total calculation time. When applied to the actual PAM system, physical filters can be used instead of digital filters, which can greatly reduce the total calculation time.

FFT time: The time required for spectrum analysis and FFT transformation. The commonly used function is x=fft(n,nfft), where nfft is the number of FFT points, which is set to half the sampling rate. From the results, we can see that without considering the filtering time (both require filtering), for the same amount of data, the processing time required for FFT transformation is more than 20 times the calculation time of this algorithm.

Example 2

This embodiment tests the acoustic data received by a buoy deployed in the ocean for one month (the recording device is SM4, the sampling rate is 500kHz, and the acquisition is collected at intervals). The click signal is detected using processing methods such as frequency band filtering, signal-to-noise ratio, and matched filtering. The accuracy of the automatic detection result is less than 30%. However, this method greatly improves the detection accuracy by processing interference signals, reaching more than 80%. It can be seen that when applied in a long-term passive monitoring system, the detection accuracy is not optimistic as the monitoring time increases.

Table 2 Accuracy test of different detection methods

Claims (2)

1. The dolphin echo positioning signal detection method applied to the PAM system is characterized by comprising the following three steps:

1) Collecting sound data, carrying out band-pass filtering on the collected sound data, and calculating by the following formula to obtain a signal-to-noise ratio (SNR):

Where s represents a time series containing signal data, sqrt (mean (s ²)) is the root mean square value RMS of the signal series, represents the noise level, and abs(s) represents the absolute value of the amplitude of the signal contained in the time series;

2) Setting a threshold value TH of signal-to-noise ratio detection, wherein the signal-to-noise ratio is 1) the numerical value of the obtained data is calculated by a formula; when the signal-to-noise ratio of the signal is satisfied

SNR＞TH

Representing the detection of suspected signal data, wherein each signal pulse on the waveform diagram consists of a group of signal data and is regarded as a suspected signal; instead, all signals less than TH are considered noise; according to the number of the detected suspected signals, setting a dynamic threshold value TH' and extracting signals again by using the above formula, wherein the calculation formula of the dynamic threshold value is as follows:

Wherein s' is equal to an average noise level mean (SNR), n is the number of suspected signal data, and t is the calculated total time length;

3) Grouping the suspected signal data higher than the dynamic threshold value, forming a complete suspected signal by each group of suspected signal data, and determining the start and stop positions of each suspected signal, wherein the method comprises the following steps of:

Setting the length dt of the detection window, wherein the value of dt is the minimum interval time between two suspected signals; storing all suspected signal data positions meeting SNR > TH into an array X, taking the first suspected signal data time in the array X as a starting point, marking time as t (1), marking the second suspected signal data time as t (2), judging the relation between t (2) -t (1) and dt, if t (2) -t (1) < dt, continuing to move to t (3) until t (n) -t (1) > dt marks t (1) to t (n-1) as a time array of the first signal.

2. The method for detecting a dolphin echo locating signal applied to a PAM system according to claim 1, wherein the suspected signal at the start-stop position determined in step 3) is selected by one or more of the following 3 methods:

(1) Length of control signal time

A. after the signal grouping, extracting time information of each suspected signal, including: calculating the time length of each suspected signal according to the starting time, the ending time and the peak time, and storing the time length as a signal group to be processed;

b. setting a time length threshold as a judging condition, and obtaining the value of the time length threshold by multiplying the average time length of all the extracted suspected signals by a coefficient, wherein the coefficient is 1.2-1.5;

c. Traversing all the suspected signals, removing the suspected signals which do not accord with the time length threshold, and storing the residual suspected signals after the removal for the next screening;

(2) Introducing a minimum time interval threshold parameter L, and meeting the condition that DeltaT is less than L and less than T;

Wherein DeltaT is the arrival time difference between the sound ray with large partial glancing angle and the direct signal after being reflected by sea surface and seabed and reaching the receiver, and T is the time interval parameter of the dolphin echo positioning signal obtained by empirical data; the value of the minimum time interval threshold L is adjusted according to the local water depth and the acoustic characteristics of the target species;

Traversing all the suspected signals, judging the time difference between the next suspected signal and the previous suspected signal from the second suspected signal, if the formula is not satisfied, eliminating the signals until the last suspected signal, and storing the residual suspected signals after elimination;

(3) Judging isolated signals

And eliminating the isolated signal from all suspected signals, wherein the isolated signal is that no suspected signal data exists or the suspected signal exists within 1 second before and after the signal, but the suspected signal does not accord with the characteristics of the dolphin echo positioning signal.