Target detection method under active sonar steady-state interference

Abstract

The invention relates to the technical field of acoustic signal processing, in particular to a target detection method under active sonar steady-state interference, and aims to solve the problem that a traditional matched filtering method is easily interfered by sonar platform self-noise under a low signal-to-noise ratio. And performing sliding windowing segmentation processing on the echo signal, calculating the KL divergence of the echo signal and the interference template, and realizing target detection by using difference characteristics. According to the method, the problem of missing detection caused by overlapping of interference and target echoes is solved, a semi-physical simulation result shows that the target can still be effectively detected when the signal-to-noise ratio is lower than-10dB, and the detection success rate is improved by more than 30%.

Description

Technical Field

[0001] This invention relates to the field of acoustic signal processing technology, specifically to a target detection method under steady-state interference of active sonar. Background Technology

[0002] Active sonar is a type of device that uses sound waves for underwater detection. It is widely used in fields such as seabed research, underwater navigation, and target detection. Its working principle is to use a sound source to emit sound wave pulses, and a receiving device to receive the reflected echoes from underwater objects. The target is detected based on the characteristics of the echo amplitude, time delay, etc.

[0003] Active sonar is often mounted on platforms such as ships to operate. In practical applications, active sonar echo detection is often interfered with by the self-noise of the platform itself or other platforms, which seriously affects the performance of the target. Platform self-noise includes mechanical vibration noise, flow noise, and propeller radiation noise. These noises include narrowband line spectrum and broadband interference components. When they are close to the sonar's operating frequency band, they are difficult to eliminate by simple filtering. Moreover, the interference is particularly prominent in the towed state, which will seriously reduce the signal-to-noise ratio of the target echo and seriously affect the sonar system's target detection performance.

[0004] Threshold detection of the peak points of echo matched filtering is a common active sonar target detection method. When the background interference intensity is large (the echo signal-to-noise ratio is low), traditional methods are difficult to achieve effective target detection. Based on this, we propose a target detection method under active sonar steady-state interference. Summary of the Invention

[0005] This invention addresses the challenge of weak target echo detection caused by steady-state interference signals such as platform self-noise during active sonar detection. It provides a target echo detection method based on KL divergence under steady-state interference, which fully utilizes the relatively stable probability distribution of steady-state interference in the matched filtering domain. By performing sliding windowing and piecewise probability density estimation on the matched filtering results, the method characterizes the distribution difference between the active sonar echo and the interference background, thereby significantly improving the weak target detection capability under interference.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a target detection method under steady-state interference of active sonar, applicable to underwater acoustic detection systems mounted on mobile or fixed platforms, comprising the following steps:

[0007] S1: Construct a steady-state interference template. By receiving steady-state interference signals from the receiving device, perform beamforming and matched filtering. Window the filtering results and estimate the probability density. Use the probability density estimation results as the steady-state interference template.

[0008] S2: Active sonar echo preprocessing, performing the same beamforming and matched filtering on the target echo signal, segmenting the signal by sliding windowing, estimating the probability density segment by segment, and constructing the echo probability density matrix;

[0009] S3: KL divergence calculation, using the interference template as a reference, calculates the relative entropy of each column of the echo probability density matrix with the interference template, and generates the difference feature vector;

[0010] S4: Constant false alarm rate target detection, which adaptively sets a threshold for the differential feature vector and achieves target detection by comparing the feature value with the threshold.

[0011] Interference template construction: Active sonar sources are turned off / paused, and signals are received only through the receiving device. The signals are then processed through downsampling, baseband reduction, and conventional beamforming to obtain beam domain data. The signal in the beam domain where the interference is located is then selected. Perform matched filtering on it to obtain the interference matched filtering result. :

[0012]

[0013] in To match the active sonar template signal, kernel density estimation is performed on each signal segment. Kernel density estimation results It can be represented as:

[0014]

[0015] in, For kernel functions, kernel estimation uses uniform kernel functions. For background statistics duration, To reduce the sampling frequency after downsampling, Kernel density estimation results As a background template for interference.

[0016] Active sonar echo preprocessing: When the active sonar is working, the echo is processed in the same way as the previous step, including downsampling, baseband reduction and conventional beamforming, to obtain the echo matched filtering domain result. Similarly, the beam where the interference is located is selected (the same beam as in the previous step). Apply a sliding window to it, and set the sliding step size to 1. , No. Segment intercept signal It can be represented as:

[0017]

[0018] Echo probability density matrix construction: Kernel density estimation is performed on each segment of the intercepted signal, and the kernel density estimation results for each segment are... Arrange the probability density functions to obtain the echo probability density matrix. :

[0019]

[0020] KL divergence calculation: Obtaining the echo probability density matrix Next, KL divergence needs to be used to characterize the difference in signal distribution features between each segment and the interference background without a target, in order to perform target detection. Therefore, the interference background template constructed in the first step can be used. Calculate separately Each column in the matrix corresponds to the interference background template. The KL divergence is used to obtain the relative entropy vector. .

[0021] Relative entropy, also known as KL divergence or information divergence, is an asymmetric measure of the difference between two probability distributions. It is defined as follows: given two probability distributions... and Relative entropy Indicates using approximate The core principle of calculating the amount of information lost is to quantify information loss by comparing the difference between two probability distributions. The formula is as follows:

[0022]

[0023] In the formula For echo probability density, This represents the probability density of the interference template.

[0024] Relative entropy is applied to vectors in this application. The calculation method can then be obtained:

[0025]

[0026] Object detection: due to the relative entropy vector The data volume is relatively small, so constant false alarm rate (CFAR) detection can be performed sequentially on each point. When performing constant false alarm rate (CFAR) detection at a given point, take that point as the center and take the values ​​before and after it. The total number of elements (excluding the point itself) is 2. A training window containing elements is used to calculate the root mean square value of the background within the training window. The calculation method is as follows:

[0027]

[0028] Obtain the root mean square value of the background Then, based on the false alarm rate requirements Based on the distribution type, determine the detection threshold. Sonar target detection often assumes that the random variable to be detected follows a normal distribution, therefore the detection threshold... Must meet:

[0029]

[0030]

[0031] In the formula As an adjustment factor used to control the detection threshold and thus meet the false alarm rate requirements, when normally distributed, it generally meets the following requirements:

[0032]

[0033] Finally With threshold The results of the active sonar target detection are compared and output:

[0034] .

[0035] Preferably, the step size of the sliding window is dynamically adjusted according to the current signal-to-noise ratio. When the signal-to-noise ratio is lower than -10dB, the step size is shortened to 50% of the original step size.

[0036] Preferably, the mobile platform includes a surface ship, an underwater robot, or an autonomous underwater vehicle.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows: Compared with the commonly used matched filter domain target detection method, this method achieves target detection by amplifying the overall probability density characteristics of the signal within a certain time period when the target signal-to-noise ratio is low and the peak amplitude is lower than the detection threshold. At the same time, it makes full use of the composition characteristics of the echo before and after the sound source is working, ensuring that there is no reverberation or target echo doping in the interference template, which can effectively solve the problem of target and interference overlap. Attached Figure Description

[0038] Figure 1 This is a flowchart of the interference suppression target detection method of the present invention;

[0039] Figure 2 This is a measured broadband warning image of active sonar echo;

[0040] Figure 3 The results of simulated echo matched filtering for a typical underwater target;

[0041] Figure 4 The results are from the simulation echo detection of steady-state interference superposition.

[0042] Figure 5 This is the target detection result based on KL divergence. Detailed Implementation

[0043] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings, so that those skilled in the art can more clearly understand how to practice the present invention. Although the present invention has been described in conjunction with its preferred embodiments, these embodiments are merely illustrative and not intended to limit the scope of the invention.

[0044] As attached Figure 1-Figure 5 As shown, a target detection method under steady-state interference of active sonar is proposed. This method mainly utilizes the difference in distribution characteristics between steady-state interference background and target echo. By estimating the probability density of the active sonar echo matched filtering result, the target is detected by quantitatively comparing the distribution differences between the echo and the interference background. In traditional active sonar echoes, it is difficult to completely separate the steady-state interference and the target echo after they are superimposed. Therefore, in order to obtain a clean interference background and a mixed signal of interference background and target echo, it is necessary to control the working state of the transmitting sound source and obtain the steady-state interference signal obtained by the signal receiving device when the active sonar sound source is not turned on, as well as the complete steady-state interference and target echo signal after the sound source is turned on. The steady-state interference signal and the active echo signal are preprocessed (baseband reduction, downsampling, beamforming, etc.) respectively.

[0045] Select the beam where the interference is located in the beam domain. Perform matched filtering on it to obtain the interference matched filtering result. and the results of matched filtering of active sonar target echoes :

[0046]

[0047] Both are subjected to matched filtering using the same matched filtering template.

[0048] The steady-state interference signal is accumulated to obtain a matched-filter time-domain signal, and the active sonar echo signal is then truncated using a sliding window. Similarly, the beam containing the interference is selected (the same beam as in the previous step). We then apply a sliding window to the signal; sliding windowing is a common segmented analysis method in signal processing, which involves gradually sliding a fixed-length window along the time sequence of the signal, extracting a segment of data within the window for processing each time; let the sliding step size be... , No. Segment intercept signal It can be represented as:

[0049]

[0050] Kernel density estimation was performed on both using kernel functions to obtain the active echo probability density functions. and the probability density function of the interference background ,Signal Kernel density estimation results It can be represented as:

[0051]

[0052]

[0053] The above formula can be used to obtain the probability density function of active echo for each segment. The kernel density estimation results of each segment Arrange the probability density functions to obtain the echo probability density matrix. :

[0054] .

[0055] During implementation, the step size of the sliding window can be adaptively adjusted based on the signal-to-noise ratio. For example, the step size of the sliding window can be dynamically adjusted according to the current signal-to-noise ratio. When the signal-to-noise ratio is lower than -10dB, the step size is shortened to 50% of the original step size.

[0056] Obtain the echo probability density matrix Next, KL divergence needs to be used to characterize the difference in signal distribution features between each segment and the interference background without a target, in order to perform target detection. Therefore, the following calculations are performed separately. Each column in the matrix corresponds to the interference background template. The KL divergence is used to obtain the relative entropy vector. The calculation method is as follows:

[0057]

[0058] Using an adaptive detection method for the relative entropy vector Detection is performed at each point, assuming that for the first point... Detection is performed at each point, and the detection background at that point is first calculated. :

[0059]

[0060] Obtain the root mean square value of the background Then, based on the false alarm rate requirements Based on the distribution type, determine the detection threshold. Sonar target detection often assumes that the random variable to be detected follows a normal distribution, therefore the detection threshold... Must meet:

[0061]

[0062]

[0063] In the formula As an adjustment factor used to control the detection threshold and thus meet the false alarm rate requirements, when normally distributed, it generally meets the following requirements:

[0064]

[0065] If the false alarm rate is set to 0.01, then Finally With threshold The results of the active sonar target detection are compared and output:

[0066]

[0067] Appendix Figure 2 The results of active sonar broadband early warning in a real marine environment are presented. There is no target echo in the figure. Strip-shaped interference appears in the red rectangular area. The interference energy is relatively concentrated in the spatial domain and does not change significantly over time. The beam domain data corresponding to the interference on the far left of the image is taken as the interference background.

[0068] Appendix Figure 3 Simulated echo matched filtering results for a typical underwater target are presented. The target is modeled and simulated using the finite element method to obtain the target response function. The echo matched filtering domain waveform is obtained by convolving the response function with the autocorrelation function of the transmitted signal. It is assumed that the target is located at 19km.

[0069] Appendix Figure 4 The results of the matched filtering domain for the simulated target superimposed interference signal are presented. The red area represents the location of the target echo, and the black dashed line represents the adaptive threshold size of the target location. As can be seen from the figure, the target intensity in the matched filtering domain does not reach the detection threshold, and the target cannot be detected. The vertical axis of the figure has been normalized and logarithmized.

[0070] Appendix Figure 5 The KL divergence output vector of the active sonar echo is given. The vertical axis of the figure has been normalized and logarithmized. Each point in the vector represents the KL divergence calculation result of the matched filter result and the interference background at that location segment. The maximum value is at 19km, indicating that the amplitude distribution of the matched filter domain in the target area is most significantly different from the interference background. The target is detected using the same threshold size. The black dashed line is the adaptive detection threshold at 19km. The figure above shows that when the traditional matched filter method cannot achieve target detection, the target detection method based on KL divergence can achieve effective target detection.

[0071] In summary, compared with commonly used matched filter domain target detection methods, this method achieves target detection by amplifying the overall probability density characteristics of the signal over a certain period of time when the target signal-to-noise ratio is low and the peak amplitude is below the detection threshold. At the same time, it makes full use of the composition characteristics of the echo before and after the sound source is working, ensuring that there is no reverberation or target echo interference in the interference template, which can effectively solve the problem of target and interference overlap.

[0072] It should also be noted that this invention is not only applicable to ship sonar, but can also be extended to: underwater robots or autonomous underwater vehicles to improve the detection capability of weak targets in autonomous navigation; seabed exploration to enhance signal resolution in complex geological environments; and underwater security to reduce false alarm rates and improve the accuracy of threat target identification.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A target detection method under steady-state interference of active sonar, characterized in that, An underwater acoustic detection system suitable for use on mobile or fixed platforms includes the following steps: S1: Construct a steady-state interference template. By receiving steady-state interference signals from the receiving device, perform beamforming and matched filtering. Window the filtering results and estimate the probability density. Use the probability density estimation results as the steady-state interference template. S2: Active sonar echo preprocessing, performing the same beamforming and matched filtering on the target echo signal, segmenting the signal by sliding windowing, estimating the probability density segment by segment, and constructing the echo probability density matrix; S3: KL divergence calculation, using the interference template as a reference, calculates the relative entropy of each column of the echo probability density matrix with the interference template, and generates the difference feature vector; S4: Constant false alarm rate target detection, which adaptively sets a threshold for the differential feature vector and achieves target detection by comparing the feature value with the threshold.

2. The target detection method under steady-state interference of active sonar according to claim 1, characterized in that, The probability density estimation in S1 employs the kernel density estimation method using a uniform kernel function, specifically as follows: The active sonar source is turned off / paused, and the signal is received only through the receiving device. The signal undergoes downsampling, baseband reduction, and conventional beamforming processing to obtain beam domain data. The signal in the beam domain where the interference is located is then selected. Perform matched filtering on it to obtain the interference matched filtering result. : in, To match the active sonar template signal, kernel density estimation is performed on each signal segment. Kernel density estimation results It can be represented as: in, For kernel functions, kernel estimation uses uniform kernel functions. For background statistics duration, To reduce the sampling frequency after downsampling, Kernel density estimation results As a background template for interference.

3. The target detection method under steady-state interference of active sonar according to claim 1, characterized in that, Specifically, S2 is as follows: When the active sonar is working, the echo is also downsampled, reduced in baseband, and processed using conventional beamforming to obtain the echo matched filtering domain result, which is then compared with the beam domain signal from step S1. Same beam signal And add a sliding window to it, setting the sliding step size to be... , No. Segment intercept signal It can be represented as: Kernel density estimation is performed on each segment of the extracted signal, and the probability density function of each segment is calculated. Arrange the array to obtain the echo probability density matrix. : 。 4. The target detection method under steady-state interference of active sonar according to claim 3, characterized in that, S3 specifically involves obtaining the echo probability density matrix. Then, the interference background template constructed in the first step is used. Calculate separately Each column in the matrix corresponds to the interference background template. The KL divergence is used to obtain the relative entropy vector. Information loss is quantified by comparing the differences between two probability distributions, and the calculation formula is as follows: In the formula For echo probability density, This represents the probability density of the interference template.

5. The target detection method under steady-state interference of active sonar according to claim 1, characterized in that, The adaptive threshold setting in S4 is based on the normal distribution assumption, and the threshold coefficient is adjusted according to the false alarm rate requirement.

6. The target detection method under active sonar steady-state interference according to claim 5, characterized in that, Specifically, S4 is as follows: For the first... When performing constant false alarm rate (CFAR) detection at a given point, take that point as the center and exclude... Its front and back The total number of elements is 2. A training window containing elements is used to calculate the root mean square value of the background within the training window. The calculation method is as follows: Obtain the root mean square value of the background Then, based on the false alarm rate requirements Based on the distribution type, determine the detection threshold. Sonar target detection often assumes that the random variable to be detected follows a normal distribution; therefore, the detection threshold... Must meet: In the formula As an adjustment factor, it must meet the following requirements: Finally With threshold The results of the active sonar target detection are compared and output: 。 7. The target detection method under steady-state interference of active sonar according to claim 1, characterized in that, The step size of the sliding window is dynamically adjusted according to the current signal-to-noise ratio. When the signal-to-noise ratio is lower than -10dB, the step size is shortened to 50% of the original step size.

8. The target detection method under steady-state interference of active sonar according to claim 1, characterized in that, The mobile platform includes surface ships, underwater robots, or autonomous underwater vehicles.

Images