Underwater target echo detection method and system based on amplitude fluctuation characteristics

Abstract

The invention provides an underwater target echo detection method and system based on amplitude fluctuation characteristics, and the method comprises the steps: S1, building a target detection model according to a preset rule; s2, acquiring an echo image, performing sliding window processing on waveform data in the echo image, and establishing a probability density function model; s3, calculating a likelihood ratio function of the waveform data according to the target detection model and the probability density function model; s4, calculating a threshold value according to the probability density function model and the likelihood ratio function; and S5, updating the echo image by using a threshold value as a criterion to obtain a target detection result. According to the invention, through constant false alarm rate detection and the Neyman-Pearson criterion, a dynamic threshold value detection method which fluctuates along with the target intensity and changes along with the background is provided, the echo image is detected, and a new echo image is formed.

Description

Technical Field

[0001] The present invention relates to the field of underwater acoustic technology, and in particular to an underwater target echo detection method and system based on amplitude fluctuation characteristics. Background Art

[0002] With the development of science and technology, the game between sonar systems and underwater targets has been escalating. By changing the geometric structure and size of the target, applying stealth materials, and improving maneuverability, a series of new underwater targets have been designed and applied in battlefield environments. As a key component of sonar, constant false alarm (CFAR) detectors are often used in important civilian and military systems. The detection performance of constant false alarm detectors has received widespread attention from the society because its detection level is directly related to whether the sonar can accurately capture the target signal. The detection performance of the constant false alarm detector is related to the noise interference in the reference windows on its left and right sides. The classic CFAR detection methods mainly include unit averaging, unit averaging selection, unit averaging selection, and sequence statistics. However, in actual environments, the non-uniform situation of clutter or reverberation, the number of interfering targets, and the position information of the reverberation edge are difficult to obtain, making target detection more difficult. At the same time, when processing multiple frames of images, due to the different echo amplitudes of the target at different angles, it is easy to cause the target to be lost under a fixed threshold. To address this problem, people proposed adaptive CFAR, which can estimate different parameters according to the environment and target, and then ensure a constant false alarm rate by adjusting the detection threshold.

[0003] The Chinese patent document with the publication number CN118191810A discloses a parameter environment adaptable constant false alarm detector based on a correlation vector machine. The detector includes: a matched filter for matching and filtering the input sonar signal or radar signal; a square detector for performing square-law detection processing on the matched filtered signal; an environment adaptation module for judging the environment in which the detection unit is located by using the model order selection method and the correlation vector machine method according to the data of the reference unit; and a parameter adaptation module for calculating the threshold value according to the environment in which the detection unit is located output by the environment adaptation module, and then judging whether the detection unit has a target. The detector of the present invention has better false alarm control capability, better environmental classification capability than VI-CFAR, and can extract the position of the reverberation edge in the reverberation edge environment; and can obtain better anti-interference capability than ACCA-CFAR. The invention adjusts the threshold value based on the complex background, but the detection effect of targets with fluctuating characteristics is poor.

[0004] From the perspective of statistical theory, in multiple frames of images, the target echo amplitude random variables have different distribution types or the same distribution type but different distribution parameters. However, in the prior art, a fixed threshold detection algorithm with a known background distribution is generally used, and its detection performance analysis and system design are mainly aimed at the traditional non-fluctuating target model. When the incoming target has distribution characteristics, the sonar system designed based on the traditional target model will mismatch the incoming target, resulting in a decrease in system performance, deterioration in detection performance, and an increase in false alarm rate. In summary, in order to solve the problem of detecting targets with fluctuating echo amplitudes in multiple frames of images, the present invention conducts research on fluctuating target model parameter estimation, theoretical performance analysis of traditional detectors to guide system design that can match system parameters of incoming targets, and detector design based on target knowledge, and provides a technical method for underwater target echo detection based on amplitude fluctuation characteristics. This method is of great significance for improving the accuracy of target models, the reliability of system design, and target detection capabilities. Summary of the invention

[0005] In view of the defects in the prior art, an object of the present invention is to provide a method and system for underwater target echo detection based on amplitude fluctuation characteristics.

[0006] According to the present invention, a method for detecting underwater target echo based on amplitude fluctuation characteristics comprises:

[0007] Step S1: Establish a target detection model according to preset rules;

[0008] Step S2: Acquire an echo image, perform sliding window processing on the waveform data in the echo image, and establish a probability density function model;

[0009] Step S3: Calculate the likelihood ratio function of the waveform data according to the target detection model and the probability density function model;

[0010] Step S4: Calculate the threshold value according to the probability density function model and the likelihood ratio function;

[0011] Step S5: Using the threshold value as a criterion, the echo image is updated to obtain the target detection result.

[0012] Preferably, step S1 comprises:

[0013] Step S1.1: preset target detection rules:

[0014]

[0015] Among them, H0 means no target, H1 means target; y(t) means echo signal, s(t) means target echo signal, c(t) means noise reverberation;

[0016] Step S1.2: Generate target probability rules according to target detection rules, and set target probability determination rules;

[0017] In H i Under the established conditions, determine H j The probability of establishment is:

[0018] P(H j |H i )=∫p(y|H i )dy

[0019] Among them, i and j both represent 0 or 1, p(y|H i ) is H i The probability density function of y under the condition, p(H j |H i ) indicates H i Assume that y satisfies H j probability;

[0020] According to the target probability rule, the decision rule is generated as follows: p(H1|H1) and p(H0|H0) are the probabilities of correct decision; p(H1|H0) is the false alarm probability, and p fa It represents the probability of judging that there is a target when the target does not exist; p(H0|H1) is the missed detection rate, which represents the probability of judging that there is no target when the target exists;

[0021] Step S1.3: Associate the target detection rules and the target probability determination rules to establish a target detection model.

[0022] Preferably, step S2 comprises:

[0023] Step S2.1: acquiring an echo image once every preset time, and dividing the echo image according to preset angles to obtain waveform data in multiple angle ranges;

[0024] Step S2.2: Select the waveform data of any angle, and combine the waveform data to form a detection unit, and process the detection unit in a sliding window form to obtain the current target detection unit;

[0025] Specifically, in the selected waveform data, the sliding window length is preset to M, M / 2 units before and after the detection unit are selected as reference units, g units adjacent to the detection unit are selected as protection units, and the detection unit, the reference unit and the protection unit are merged as the target detection unit for this time;

[0026] Step S2.3: Based on the target detection model, set the probability density calculation rules, and calculate the parameters of this detection through the preset rules using the echo data of the target detection unit, and establish the probability density model;

[0027] Specifically,

[0028] The probability density model of noise reverberation is:

[0029]

[0030] c represents the random variable parameter, e is the preset parameter, where the noise satisfies the Gaussian distribution, the mean is U, and the variance is σ 2 ;

[0031] The probability density model of the target echo signal is:

[0032]

[0033] Among them, the amplitude fluctuation of the target echo s(t) follows the Ts distribution, and the parameter k is the degree of freedom, is the average target intensity, Γ(·) is the gamma function, which is:

[0034] Γ(x)=∫0 +∞ t x-1 e -t dt(x>0)

[0035] t is the integration variable.

[0036] Preferably, step S3 includes:

[0037] Step S3.1: further process the echo data of the target detection unit of this time, and calculate the probability density of the echo signal y of the target detection unit of this time under the conditions of H0 and H1 respectively;

[0038] Under the H0 condition:

[0039]

[0040] Under H1 conditions:

[0041]

[0042] Wherein, f(c) is the probability density function of the noise, g(yc) is the probability density of the target echo amplitude fluctuation, step S3.2: calculate the likelihood ratio function value:

[0043] λ(y)=p(y|H1) / p(y|H0).

[0044] Preferably, step S4 comprises:

[0045] Step S4.1: Obtain the echo signal z of each reference unit, calculate the corresponding λ(z) value, and calculate the probability density function p(λ(z)|H0) of λ(z) under the condition of H0;

[0046] Step S4.2: Calculate the threshold μ according to the calculated parameters of the echo signal z, which is specifically determined by the following formula:

[0047]

[0048] Wherein, α is a preset false alarm probability value, the threshold μ is the lower limit of the integral function, μ ≥ 0 and is adaptive.

[0049] Preferably, step S5 comprises:

[0050] Step S5.1: Determine the probability function value according to the Neyman-Pearson criterion and classify the likelihood ratio function as follows:

[0051]

[0052] Step S5.2: Compare the likelihood ratio function value with the threshold value. When the likelihood ratio function is greater than μ, retain the echo signal corresponding to the target detection unit this time. When the likelihood ratio function is less than μ, assign a uniform preset value to the corresponding echo signal. Thus, the target detection result is obtained.

[0053] Step S5.3: The target detection result is formed into a detection echo image, the sliding window position is changed, and steps S2 to S5 are repeated until all target detection units are detected, and the final detected echo image is output.

[0054] According to the present invention, an underwater target echo detection system based on amplitude fluctuation characteristics includes:

[0055] Module M1: Establish target detection model according to preset rules;

[0056] Module M2: Acquire the echo image, perform sliding window processing on the waveform data in the echo image, and establish a probability density function model;

[0057] Module M3: Calculate the likelihood ratio function of the waveform data based on the target detection model and the probability density function model;

[0058] Module M4: Calculate the threshold value according to the probability density function model and the likelihood ratio function;

[0059] Module M5: Using the threshold as a criterion, the echo image is updated to obtain the target detection result.

[0060] Preferably, module M1 comprises:

[0061] Module M1.1: preset target detection rules are:

[0062]

[0063] Among them, H0 means no target, H1 means target; y(t) means echo signal, s(t) means target echo signal, c(t) means noise reverberation;

[0064] Module M1.2: Generate target probability rules based on target detection rules, and set target probability determination rules;

[0065] In H i Under the established conditions, determine H j The probability of establishment is:

[0066] P(H j |H i )=∫p(y|h i )dy

[0067] Among them, i and j both represent 0 or 1, p(y|H i ) is H i The probability density function of y under the condition, p(H j |H i ) indicates H i Assume that y satisfies H j probability;

[0068] According to the target probability rule, the decision rule is generated as follows: p(H1|H1) and p(H0|H0) are the probabilities of correct decision; p(H1|H0) is the false alarm probability, and p fa It represents the probability of judging that there is a target when the target does not exist; p(H0|H1) is the missed detection rate, which represents the probability of judging that there is no target when the target exists;

[0069] Module M1.3: Associate target detection rules and target probability judgment rules to establish a target detection model;

[0070] Module M2 includes:

[0071] Module M2.1: Acquire an echo image at a preset time interval, and divide the echo image according to preset angles to obtain waveform data in multiple angle ranges;

[0072] Module M2.2: Select the waveform data of any angle, combine the waveform data to form a detection unit, process the detection unit in a sliding window form, and obtain the target detection unit for this time;

[0073] Specifically, in the selected waveform data, the sliding window length is preset to M, M / 2 units before and after the detection unit are selected as reference units, g units adjacent to the detection unit are selected as protection units, and the detection unit, the reference unit and the protection unit are merged as the target detection unit for this time;

[0074] Module M2.3: Based on the target detection model, set the probability density calculation rules, and calculate the parameters of this detection through the preset rules using the echo data of the target detection unit to establish the probability density model;

[0075] Specifically,

[0076] The probability density model of noise reverberation is:

[0077]

[0078] c represents the random variable parameter, e is the preset parameter, where the noise satisfies the Gaussian distribution, the mean is U, and the variance is σ 2 ;

[0079] The probability density model of the target echo signal is:

[0080]

[0081] Among them, the amplitude fluctuation of the target echo s(t) follows the Ts distribution, and the parameter k is the degree of freedom, is the average target intensity, Γ(·) is the gamma function, which is:

[0082] Γ(x)=∫0 +∞ t x-1 e -t dt(x>0)

[0083] t is the integration variable.

[0084] Preferably, module M3 comprises:

[0085] Module M3.1: further process the echo data of the target detection unit, and calculate the probability density of the echo signal y of the target detection unit under the conditions of H0 and H1 respectively;

[0086] Under the H0 condition:

[0087]

[0088] Under H1 conditions:

[0089]

[0090] Among them, f(c) is the probability density function of noise, g(yc) is the probability density of target echo amplitude fluctuation,

[0091] Module M3.2: Calculate the likelihood ratio function value:

[0092] λ(y)=p(y|H1) / p(y|H0);

[0093] Module M4 includes:

[0094] Module M4.1: Obtain the echo signal z of each reference unit, calculate the corresponding λ(z) value, and calculate the probability density function p(λ(z)|H0) of λ(z) under the condition of H0;

[0095] Module M4.2: Calculate the threshold μ based on the calculated parameters of the echo signal z, which is specifically determined by the following formula:

[0096]

[0097] Wherein, α is a preset false alarm probability value, the threshold μ is the lower limit of the integral function, μ ≥ 0 and is adaptive.

[0098] Preferably, module M5 comprises:

[0099] Module M5.1: Determine the value of the probability function according to the Neyman-Pearson criterion and classify the likelihood ratio function into

[0100]

[0101] Module M5.2: Compare the likelihood ratio function value with the threshold value. When the likelihood ratio function is greater than μ, retain the echo signal corresponding to the target detection unit. When the likelihood ratio function is less than μ, assign a uniform preset value to the corresponding echo signal. Thus, the target detection result is obtained.

[0102] Module M5.3: The target detection result is formed into a detection echo image, the sliding window position is changed, and the module M2-module M5 are repeatedly triggered until all target detection units are detected, and the final detected echo image is output.

[0103] Compared with the prior art, the present invention has the following beneficial effects:

[0104] 1. The present invention provides a dynamic threshold detection method that changes with target intensity fluctuations and background changes through constant false alarm detection and Neyman-Pearson criterion. The effectiveness of the method is proved by detecting echo images. In the present invention, the threshold value changes continuously, which can ensure that the false alarm rate during detection is constant.

[0105] 2. The target detection model based on amplitude fluctuation characteristics can more accurately identify the difference between underwater target echo signals and noise reverberation. By using the likelihood ratio function and the threshold value as the detection criteria, the false alarm probability and missed detection rate can be significantly reduced, thereby improving the accuracy of target detection.

[0106] 3. The threshold value set by the present invention is adaptive and can be dynamically adjusted according to the echo signal parameters of each reference unit, so as to adapt to the target detection requirements under different environments and conditions. This adaptive performance enables the method to maintain high detection performance in various underwater environments;

[0107] 4. The present invention can efficiently process a large amount of echo data and quickly identify potential target echo signals through sliding window processing and the establishment of a probability density function model. In addition, the present invention can also update the detection results to the echo image in real time, which is convenient for subsequent analysis and processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0109] Figure 1 It is a flow chart of the underwater target echo detection method based on amplitude fluctuation characteristics of the present invention;

[0110] Figure 2 is a flow chart of the method of embodiment 1 of the present invention;

[0111] Figure 3 This is a schematic diagram of the sliding window described in Example 1 of the present invention;

[0112] Figure 4 The simulation detection image comparison described in Example 2 of the present invention;

[0113] Figure 5 This is the actual data detection image comparison described in Example 2 of the present invention. DETAILED DESCRIPTION

[0114] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0115] In order to study the underwater target echo detection technology method with amplitude fluctuation characteristics in the above-mentioned background technology, the present invention provides an echo detection technology method in which the background obeys a Gaussian distribution model and the target amplitude fluctuation obeys a Ts distribution (derived from the chi-square distribution). The time domain waveform is divided into a reference unit, a detection unit and a protection unit by a sliding window method. The reference unit is used to estimate the background model parameters, and then the detection unit is combined to perform adaptive threshold calculation. Finally, the Neyman-Pearson criterion is used to make a judgment and give a final result. The method flow chart is shown in FIG. Figure 2This method takes underwater targets with amplitude fluctuation characteristics as the research object, adjusts the threshold through the estimation parameters of different backgrounds, thereby ensuring a constant false alarm rate, making it more reliable and adaptive in target detection, and can provide a reference for improving the detection capability of sonar.

[0116] According to the present invention, a method for detecting underwater target echo based on amplitude fluctuation characteristics comprises:

[0117] Step S1: Establish a target detection model according to preset rules; Step S2: Acquire an echo image, perform sliding window processing on the waveform data in the echo image, and establish a probability density function model; Step S3: Calculate the likelihood ratio function of the waveform data according to the target detection model and the probability density function model; Step S4: Calculate the threshold value according to the probability density function model and the likelihood ratio function; Step S5: Use the threshold value as a criterion to update the echo image and obtain the target detection result.

[0118] Specifically, step S1 includes: Step S1.1: preset target detection rules, which are:

[0119]

[0120] Among them, H0 means no target, H1 means target; y(t) means echo signal, s(t) means target echo signal, c(t) means noise reverberation;

[0121] Step S1.2: Generate target probability rules according to target detection rules, and set target probability determination rules;

[0122] In H i Under the established conditions, determine H j The probability of establishment is:

[0123] P(H j |H i )=∫p(y|H i )dy

[0124] Among them, i and j both represent 0 or 1, p(y|H i ) is H i The probability density function of y under the condition, p(H j |H i ) indicates H i Assume that y satisfies H j probability;

[0125] According to the target probability rule, the decision rule is generated as follows: p(H1|H1) and p(H0|H0) are the probabilities of correct decision; p(H1|H0) is the false alarm probability, and p faIt represents the probability of judging that there is a target when the target does not exist; p(H0|H1) is the missed detection rate, which represents the probability of judging that there is no target when the target exists;

[0126] Step S1.3: Associate the target detection rules and the target probability determination rules to establish a target detection model.

[0127] Specifically, step S2 includes: step S2.1: acquiring an echo image once at a preset time interval, and dividing the echo image according to a preset angle to obtain waveform data of multiple angle ranges; step S2.2: selecting waveform data of any angle, and combining the waveform data to form a detection unit, and processing the detection unit in a sliding window form to obtain the target detection unit for this time;

[0128] Specifically, in the selected waveform data, the sliding window length is preset to M, M / 2 units before and after the detection unit are selected as reference units, g units adjacent to the detection unit are selected as protection units, and the detection unit, the reference unit and the protection unit are merged as the target detection unit for this time; Step S2.3: Based on the target detection model, the probability density calculation rule is set, and the echo data of the target detection unit for this time is calculated according to the preset rule to calculate the parameters of this detection, and the probability density model is established; Specifically,

[0129] The probability density model of noise reverberation is:

[0130]

[0131] c represents the random variable parameter, e is the preset parameter, where the noise satisfies the Gaussian distribution, the mean is U, and the variance is σ 2 ;

[0132] The probability density model of the target echo signal is:

[0133]

[0134] Among them, the amplitude fluctuation of the target echo s(t) follows the Ts distribution, and the parameter k is the degree of freedom, is the average target intensity, Γ(·) is the gamma function, which is:

[0135] Γ(x)=∫0 +∞ t x-1 e -t dt(x>0)

[0136] t is the integration variable.

[0137] Specifically, step S3 includes: step S3.1: further processing the echo data of the target detection unit of this time, and calculating the probability density of the echo signal y of the target detection unit of this time under the conditions of H0 and H1 respectively;

[0138] Under the H0 condition:

[0139]

[0140] Under H1 conditions:

[0141]

[0142] Wherein, f(c) is the probability density function of the noise, g(yc) is the probability density of the target echo amplitude fluctuation, step S3.2: calculate the likelihood ratio function value:

[0143] λ(y)=p(y|H1) / p(y|H0).

[0144] Specifically, step S4 includes:

[0145] Step S4.1: Obtain the echo signal z of each reference unit, calculate the corresponding λ(z) value, and calculate the probability density function p(λ(z)|H0) of λ(z) under the condition of H0;

[0146] Step S4.2: Calculate the threshold μ according to the calculated parameters of the echo signal z, which is specifically determined by the following formula:

[0147]

[0148] Wherein, α is a preset false alarm probability value, the threshold μ is the lower limit of the integral function, μ ≥ 0 and is adaptive.

[0149] Specifically, step S5 includes:

[0150] Step S5.1: Determine the probability function value according to the Neyman-Pearson criterion and classify the likelihood ratio function as follows:

[0151]

[0152] Step S5.2: Compare the likelihood ratio function value with the threshold value. When the likelihood ratio function is greater than μ, retain the echo signal corresponding to the target detection unit. When the likelihood ratio function is less than μ, assign a uniform preset value to the corresponding echo signal. For example, a minimum value may be selected. When generating the echo image, the preset value is used as the background parameter. Thus, the target detection result is obtained.

[0153] Step S5.3: The target detection result is formed into a detection echo image, the sliding window position is changed, and steps S2 to S5 are repeated until all target detection units are detected, and the final detected echo image is output.

[0154] Example 1

[0155] In order to achieve the above object, the present invention adopts the following technical solutions:

[0156] Step 1: Build an object detection model

[0157] Assuming that the sonar system receives the echo signal, after frequency conversion, demodulation, matched filtering, sampling and other processing, the target detection problem can be written as the following binary hypothesis testing problem:

[0158]

[0159] In the binary hypothesis test process, H0 indicates that there is no target, and H1 indicates that there is a target. If y(t) represents the echo signal, s(t) represents the target echo signal, and c(t) represents the reverberation, then when there is no target in the detection unit, the echo only has reverberation, and when there is a target, the received signal consists of the target echo and reverberation.

[0160] Because the observed samples are random, signal detection actually uses statistical methods to process random variables, so it is described using statistical probability:

[0161] In H i Under the assumption that H j The probability of establishment is:

[0162] P(H j |H i )=∫p(y|H i )dy (2)

[0163] Among them, p(y|H i ) is H i Assume that the probability density function of y is true, p(H j |H i ) indicates H i Assume that y satisfies H j Under the binary hypothesis, p(H j |H i ) has 4 results: p(H1|H1) (detection probability, generally p d denoted) and p(H0|H0) denote the probability of correct judgment, and p(H1|H0) is called the false alarm probability, generally expressed as p faIt represents the probability of judging that there is a target when the target does not exist, and p(H0|H1) is called the missed detection rate, which represents the probability of judging that there is no target when the target exists. In summary, we get the target detection model.

[0164] Step 2: Get the detection image and slide window processing

[0165] Read a frame of echo image, divide it according to the time domain waveform of different angles, and process it in the form of sliding window or extract a whole time domain waveform. The processing process is as follows: in the waveform data of the selected angle, set the sliding window length to M, select M / 2 reference units before and after the detection unit, and in order to prevent the signal of the detection unit from leaking to the adjacent reference unit, the adjacent g units need to be used as protection units, so as to ensure the effective estimation of the background power. The schematic diagram is as follows Figure 3 .

[0166] Based on the detection model, the present invention considers that the background reverberation c(t) satisfies Gaussian distribution, independent and identically distributed, with mean U and variance σ 2 , the corresponding probability density (pdf) is:

[0167]

[0168] c represents a random variable, and the noise reverberation c(t) satisfies the Gaussian distribution, with a mean of U and a variance of σ 2 ,

[0169] The parameters are determined by fitting the data in the actual sliding window after removing the detection unit and the protection unit. The parameters are different under different sliding windows.

[0170] The amplitude fluctuation of the target echo s(t) follows the Ts distribution, and the parameters are k is the degree of freedom, is the average target intensity, and its probability density function is:

[0171]

[0172] Where Γ(·) is the gamma function, which is defined in the real number domain as:

[0173]

[0174] The parameter values ​​are calculated from historical multi-frame image data statistics or given by experts.

[0175] In summary, the Gaussian background fitting parameters of the echo data in the sliding window and the parameters of the target echo fluctuation model are obtained.

[0176] Step 3: Calculate the likelihood ratio function

[0177] According to step 2, under the condition of H0, the echo signal y satisfies:

[0178]

[0179] For p(y|H1), since the noise and target echo are independent of each other, the probability density function of the sum of the two in the detection model can be calculated using the convolution formula:

[0180]

[0181] Among them, f(c) is the probability density function of the noise, which is obtained by formula (3) in step 2, and g(yc) is the probability density of the target echo amplitude fluctuation, which is obtained by formula (4) in step 2. In summary, the likelihood ratio function value can be calculated:

[0182] λ(y)=p(y|H1) / p(y|H0) (8)

[0183] The main principles of this step are:

[0184] The present invention mainly adopts the Neyman-Pearson criterion for judgment, wherein the Neyman-Pearson criterion refers to ensuring the false alarm rate p of signal detection. fa =α, minimize p(H0|H1), and accordingly, minimize p d =1-p(H0|H1)=βmaximum.

[0185] Whether this criterion has an optimal solution, the decision rule is to divide the observation sample space into R0 and R1. In principle, there are infinite ways to divide the decision domain that can satisfy p(H0|H1)=α, and each method corresponds to a β. Since the likelihood function is bounded, β is a function of the partitioning method, and it is continuously bounded in the closed set, and there must be an extreme value, that is, there must be a partitioning method that maximizes β, so the solution to the Neyman-Pearson criterion exists.

[0186] For this constrained optimization problem, the Lagrange multiplier method is used to construct the objective function J and express it in integral form, namely:

[0187] J=p(H0|H1)+μ[p(H1|H0)-α]=∫ R0 p(y|H1)dy+μ[∫ R0 p(y|H0)dy-α] (9)

[0188] Among them, μ≥0, is the Lagrange multiplier. The above formula can be resolved as:

[0189] J=μ(1-α)+∫ R0 [p(y|H1)-μp(y|H0)]dy (10)

[0190] It can be seen that the first term in the above formula is positive, so the negative part of the integral term is assigned to the R0 region, and the rest is assigned to the R1 region, which can ensure that J is minimum, and we have:

[0191] H0: p(y|H1) <p(y|H0)

[0192] H1: p(y|H1)>p(y|H0) (11)

[0193] Written in the form of a likelihood ratio test:

[0194]

[0195] The above formula gives the solution to the constrained optimization problem, that is, the part of the likelihood ratio function greater than μ is divided into the R1 region, and the rest is divided into the R0 region. In the formula, μ is equivalent to the threshold value, which is calculated and determined by step 4.

[0196] Step 4: Calculate the threshold

[0197] For each reference unit value z in the sliding window, the corresponding λ(z) value is calculated, and then the parameters (U1,σ1) are determined by fitting using equation (3), and the probability density function p(λ(z)|H0) of λ(z) under the H0 hypothesis is obtained. The threshold μ is then determined by the following equation:

[0198]

[0199] Among them, α is a given false alarm probability value, and μ is the lower limit of the integral in the formula, which is adaptive.

[0200] For a detection environment with constant background power, p(λ(z)|H0) is fixed and constant, and CFAR detection can be achieved by calculating μ using the above formula. If the background power varies with time, p(λ(z)|H0) will change accordingly. In this case, a CFAR processing method is needed to dynamically estimate the background power level so that the threshold μ is adaptive, that is, to provide an integral lower limit that changes with p(λ(z)|H0) for the above formula, thereby ensuring that the above formula always holds.

[0201] Step 5: Image Detection

[0202] For the value y of the detection unit, the corresponding value of λ(y) can be obtained, and then compared with the corresponding threshold μ. If it is greater than, it is judged as a target, and the value of the detection unit is retained in the final image. If it is less than, it is judged as no target, and the detection unit is assigned a minimum value in the final image. Then it is determined whether all units have been detected. If there are still undetected units, change the sliding window position (move one unit to the next moment) and continue from step 3. If there are no undetected units, output the final detected image.

[0203] The theoretical basis and working principle of the present invention are:

[0204] (1) Constant false alarm detection: It is an adaptive threshold detection algorithm that aims to obtain predictable detection probability and false alarm probability in an interference environment. It is widely used in the field of automatic detection of radar and sonar signals. Its function is to make automatic detection free from the influence of background interference power fluctuations and maintain a constant false alarm probability. Its basic principle is to use the auxiliary data of the reference unit to estimate the background interference power of the detection unit, so as to obtain a detection threshold that is adaptive to changes in the background environment.

[0205] (2) Neyman-Pearson criterion: Under the condition of a given false alarm probability, the detection probability is maximized by adjusting the detection statistic.

[0206] The present invention performs target detection on a single-frame image based on constant false alarm detection and the Neyman-Pearson criterion. First, the echo image is divided according to the time domain, and a sliding window is used to slide on the time domain signal to extract the reference unit, the detection unit and the protection unit; then, the reference unit is statistically analyzed to fit the corresponding normal distribution parameters; then, the likelihood ratio function is calculated in combination with the estimated target distribution parameters, and finally, the threshold value is estimated and compared in turn to obtain the final image after detection, thereby providing a detection method in which the threshold value changes with the background, which can provide a certain theoretical basis for target detection.

[0207] Example 2

[0208] (1) Data reading

[0209] Simulation image: Set the signal-to-noise ratio to 10-20dB, the signal to 1, and the noise to additive Gaussian white noise satisfying N(0,1). Randomly generate a set of 800*20 data, the image is as follows Figure 4 (a) shown.

[0210] Actual data image: According to the acquisition results, 8138*288 data is obtained, where 288 corresponds to 360 degrees and 8138 corresponds to a distance of 500 meters. The corresponding image is as follows: Figure 5 (a) shown.

[0211] (2) Fitting distribution parameters through reference units

[0212] Simulation image: Set the window length to 800 and the protection unit length to 1, that is, take the entire time domain signal for processing at one time. The average of the data excluding the detection unit and the protection unit is the estimated normal distribution U value, and then calculate the standard deviation σ.

[0213] Actual data image: Set the window length to 300 and the protection unit length to 1, that is, take 300 reference units before and after the detection unit as background estimation. Then take the average of the data excluding the detection unit and the protection unit to get the estimated normal distribution U value, and then calculate the standard deviation σ.

[0214] (3) Likelihood ratio function calculation

[0215] Assuming the value of the detection unit is y, the denominator of the likelihood ratio function is:

[0216]

[0217] The calculation method of numerator p(y|H1) is as follows:

[0218]

[0219] Among them, the distribution parameters of the target echo are Based on the existing multi-frame image data, the value is (0.2, -30). In summary, the likelihood ratio function can be calculated.

[0220] (4) Threshold calculation and comparison

[0221] Assuming the false alarm rate α = 0.01, the threshold threshold can be calculated by equation (13) and the fitting parameters (U1, σ1). Then the likelihood ratio function λ(y) corresponding to the detection unit y is compared with the threshold. If it is greater than the threshold, the detection unit value is retained, and if it is less than the threshold, a minimum value is assigned.

[0222] Repeat the above steps until all units have been tested. Figure 4 (b) is the result of the simulation data after being tested by this algorithm. Figure 5 (b) is the result of the test data after being detected by this algorithm. It can be seen from the result that the target features are retained, indicating that this detection method is effective.

[0223] The present invention also provides an underwater target echo detection system based on amplitude fluctuation characteristics. The underwater target echo detection system based on amplitude fluctuation characteristics can be realized by executing the process steps of the underwater target echo detection method based on amplitude fluctuation characteristics, that is, those skilled in the art can understand the underwater target echo detection method based on amplitude fluctuation characteristics as a preferred implementation of the underwater target echo detection system based on amplitude fluctuation characteristics.

[0224] According to the present invention, an underwater target echo detection system based on amplitude fluctuation characteristics includes: module M1: establishing a target detection model according to preset rules; module M2: acquiring an echo image, performing sliding window processing on waveform data in the echo image, and establishing a probability density function model; module M3: calculating a likelihood ratio function of the waveform data according to the target detection model and the probability density function model; module M4: calculating a threshold value according to the probability density function model and the likelihood ratio function; module M5: using the threshold value as a criterion, updating the echo image to obtain a target detection result.

[0225] Specifically, module M1 includes: Module M1.1: preset target detection rules, which are:

[0226]

[0227] Among them, H0 means no target, H1 means target; y(t) means echo signal, s(t) means target echo signal, c(t) means noise reverberation; Module M1.2: Generate target probability rule according to target detection rule, and set target probability judgment rule at the same time; i Under the established conditions, determine H j The probability of establishment is:

[0228] P(H j |H i )=∫p(y|H i )dy

[0229] Among them, i and j both represent 0 or 1, p(y|H i ) is H i The probability density function of y under the condition, p(H j |H i ) indicates H i Assume that y satisfies H j According to the target probability rule, the decision rule is generated as follows: p(H1|H1) and p(H0|H0) are the probabilities of correct decision; p(H1|H0) is the false alarm probability, and p fa represents the probability of judging that there is a target when the target does not exist; p(H0|H1) is the missed detection rate, which represents the probability of judging that there is no target when the target exists; Module M1.3: Associate the target detection rules and the target probability judgment rules to establish a target detection model;

[0230] Module M2 includes: Module M2.1: Acquire an echo image at a preset time interval, and divide the echo image according to preset angles to obtain waveform data of multiple angle ranges; Module M2.2: Select waveform data of any angle, and combine the waveform data to form a detection unit, and process the detection unit in a sliding window form to obtain the current target detection unit; Specifically, in the selected waveform data, the preset sliding window length is M, and M / 2 units before and after the detection unit are selected as reference units, and g units adjacent to the detection unit are selected as protection units, and the detection unit, the reference unit and the protection unit are merged as the current target detection unit; Module M2.3: Based on the target detection model, set the probability density calculation rule, and calculate the parameters of this detection through the preset rules of the echo data of the current target detection unit, and establish a probability density model; Specifically,

[0231] The probability density model of noise reverberation is:

[0232]

[0233] c represents the random variable parameter, e is the preset parameter, where the noise satisfies the Gaussian distribution, the mean is U, and the variance is σ 2 ;

[0234] The probability density model of the target echo signal is:

[0235]

[0236] Among them, the amplitude fluctuation of the target echo s(t) follows the Ts distribution, and the parameter k is the degree of freedom, is the average target intensity, Γ(·) is the gamma function, which is:

[0237] Γ(x)=∫0 +∞ t x-1 e -t dt(x>0)

[0238] t is the integration variable.

[0239] Specifically, module M3 includes: module M3.1: further processing the echo data of the target detection unit of this time, and calculating the probability density of the echo signal y of the target detection unit of this time under the conditions of H0 and H1 respectively;

[0240] Under the H0 condition:

[0241]

[0242] Under H1 conditions:

[0243]

[0244] Where f(c) is the probability density function of the noise, g(yc) is the probability density of the target echo amplitude fluctuation, and module M3.2: calculates the likelihood ratio function value:

[0245] λ(y)=p(y|H1) / p(y|H0);

[0246] Module M4 includes: Module M4.1: Obtain the echo signal z of each reference unit, calculate the corresponding λ(z) value, and calculate the probability density function p(λ(z)|H0) of λ(z) under the condition of H0; Module M4.2: Calculate the threshold μ according to the calculated parameters of the echo signal z, which is specifically determined by the following formula:

[0247]

[0248] Wherein, α is a preset false alarm probability value, the threshold μ is the lower limit of the integral function, μ ≥ 0 and is adaptive.

[0249] Specifically, module M5 includes: Module M5.1: Determine the probability function value according to the Neyman-Pearson criterion and classify the likelihood ratio function into

[0250]

[0251] Module M5.2: Compare the likelihood ratio function value with the threshold value. When the likelihood ratio function is greater than μ, retain the echo signal corresponding to the target detection unit. When the likelihood ratio function is less than μ, assign a uniform preset value to the corresponding echo signal, thereby obtaining the target detection result. Module M5.3: Form the target detection result into a detection echo image, change the sliding window position, and repeatedly trigger module M2-module M5 until all target detection units have been detected, and output the final detected echo image.

[0252] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0253] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for detecting underwater target echo based on amplitude fluctuation characteristics, characterized in that: include: Step S1: Establish a target detection model according to preset rules; Step S2: Acquire an echo image, perform sliding window processing on the waveform data in the echo image, and establish a probability density function model; Step S3: Calculate the likelihood ratio function of the waveform data according to the target detection model and the probability density function model; Step S4: Calculate the threshold value according to the probability density function model and the likelihood ratio function; Step S5: Using the threshold value as a criterion, the echo image is updated to obtain the target detection result.

2. The underwater target echo detection method based on amplitude fluctuation characteristics according to claim 1 is characterized in that: Step S1 includes: Step S1.1: preset target detection rules: Among them, H0 means no target, H1 means target; y(t) means echo signal, s(t) means target echo signal, c(t) means noise reverberation; Step S1.2: Generate target probability rules according to target detection rules, and set target probability determination rules; In H i Under the established conditions, determine H j The probability of establishment is: P(H j |H i )=∫p(y|H i )of Among them, i and j both represent 0 or 1, p(y|H i ) is H i The probability density function of y under the condition, p(H j |H i ) indicates H i Assume that y satisfies H j probability; According to the target probability rule, the decision rule is generated as follows: p(H1|H1) and p(H0|H0) are the probabilities of correct decision; p(H1|H0) is the false alarm probability, and p fa It represents the probability of judging that there is a target when there is no target; p(H0|H1) is the missed detection rate, which represents the probability of judging that there is no target when there is a target; Step S1.3: Associate the target detection rules and the target probability determination rules to establish a target detection model.

3. The underwater target echo detection method based on amplitude fluctuation characteristics according to claim 1 is characterized in that: Step S2 includes: Step S2.1: acquiring an echo image once every preset time, and dividing the echo image according to preset angles to obtain waveform data in multiple angle ranges; Step S2.2: Select the waveform data of any angle, and combine the waveform data to form a detection unit, and process the detection unit in a sliding window form to obtain the current target detection unit; Specifically, in the selected waveform data, the sliding window length is preset to M, M / 2 units before and after the detection unit are selected as reference units, g units adjacent to the detection unit are selected as protection units, and the detection unit, the reference unit and the protection unit are merged as the target detection unit for this time; Step S2.3: Based on the target detection model, set the probability density calculation rules, and calculate the parameters of this detection through the preset rules using the echo data of the target detection unit, and establish the probability density model; Specifically, The probability density model of noise reverberation is: c represents the random variable parameter, e is the preset parameter, where the noise satisfies the Gaussian distribution, the mean is U, and the variance is σ 2 ; The probability density model of the target echo signal is: Among them, the amplitude fluctuation of the target echo s(t) follows the Ts distribution, and the parameter k is the degree of freedom, is the average target intensity, Γ(·) is the gamma function, which is: Γ(x)=∫0 +∞ t x-1 e -t dt(x>0) t is the integration variable.

4. The underwater target echo detection method based on amplitude fluctuation characteristics according to claim 1 is characterized in that: Step S3 includes: Step S3.1: further process the echo data of the target detection unit of this time, and calculate the probability density of the echo signal y of the target detection unit of this time under the conditions of H0 and H1 respectively; Under the H0 condition: Under H1 conditions: Among them, f(c) is the probability density function of noise, g(yc) is the probability density of target echo amplitude fluctuation, Step S3.2: Calculate the likelihood ratio function value: λ(y)=p(y|H1) / p(y|H0).

5. The underwater target echo detection method based on amplitude fluctuation characteristics according to claim 1, characterized in that: Step S4 includes: Step S4.1: Obtain the echo signal z of each reference unit, calculate the corresponding λ(z) value, and calculate the probability density function p(λ(z)|H0) of λ(z) under the condition of H0; Step S4.2: Calculate the threshold μ according to the calculated parameters of the echo signal z, which is specifically determined by the following formula: P fa =∫ μ ∞ p(λ(z)|H0)dλ(z)=α Wherein, α is a preset false alarm probability value, the threshold μ is the lower limit of the integral function, μ ≥ 0 and is adaptive.

6. The underwater target echo detection method based on amplitude fluctuation characteristics according to claim 1, characterized in that: Step S5 includes: Step S5.1: Determine the probability function value according to the Neyman-Pearson criterion and classify the likelihood ratio function as follows: Step S5.2: Compare the likelihood ratio function value with the threshold value. When the likelihood ratio function is greater than μ, retain the echo signal corresponding to the target detection unit this time. When the likelihood ratio function is less than μ, assign a uniform preset value to the corresponding echo signal. Thus, the target detection result is obtained. Step S5.3: The target detection result is formed into a detection echo image, the sliding window position is changed, and steps S2 to S5 are repeated until all target detection units are detected, and the final detected echo image is output.

7. An underwater target echo detection system based on amplitude fluctuation characteristics, characterized in that: include: Module M1: Establish target detection model according to preset rules; Module M2: Acquire the echo image, perform sliding window processing on the waveform data in the echo image, and establish a probability density function model; Module M3: Calculate the likelihood ratio function of the waveform data based on the target detection model and the probability density function model; Module M4: Calculate the threshold value according to the probability density function model and the likelihood ratio function; Module M5: Using the threshold as a criterion, the echo image is updated to obtain the target detection result.

8. The underwater target echo detection system based on amplitude fluctuation characteristics according to claim 7, characterized in that: Module M1 includes: Module M1.1: preset target detection rules are: Among them, H0 means no target, H1 means target; y(t) means echo signal, s(t) means target echo signal, c(t) means noise reverberation; Module M1.2: Generate target probability rules based on target detection rules, and set target probability determination rules; In H i Under the established conditions, determine H j The probability of establishment is: P(H j |H i )=∫p(y|H i )of Among them, i and j both represent 0 or 1, p(y|H i ) is H i The probability density function of y under the condition, p(H j |H i ) indicates H i Assume that y satisfies H j probability; According to the target probability rule, the decision rule is generated as follows: p(H1|H1) and p(H0|H0) are the probabilities of correct decision; p(H1|H0) is the false alarm probability, and p fa It represents the probability of judging that there is a target when there is no target; p(H0|H1) is the missed detection rate, which represents the probability of judging that there is no target when there is a target; Module M1.3: Associate target detection rules and target probability judgment rules to establish a target detection model; Module M2 includes: Module M2.1: Acquire an echo image at a preset time interval, and divide the echo image according to preset angles to obtain waveform data in multiple angle ranges; Module M2.2: Select the waveform data of any angle, combine the waveform data to form a detection unit, process the detection unit in a sliding window form, and obtain the target detection unit for this time; Specifically, in the selected waveform data, the sliding window length is preset to M, M / 2 units before and after the detection unit are selected as reference units, g units adjacent to the detection unit are selected as protection units, and the detection unit, the reference unit and the protection unit are merged as the target detection unit for this time; Module M2.3: Based on the target detection model, set the probability density calculation rules, and calculate the parameters of this detection through the preset rules using the echo data of the target detection unit to establish the probability density model; Specifically, The probability density model of noise reverberation is: c represents the random variable parameter, e is the preset parameter, where the noise satisfies the Gaussian distribution, the mean is U, and the variance is σ 2 ; The probability density model of the target echo signal is: Among them, the amplitude fluctuation of the target echo s(t) follows the Ts distribution, and the parameter k is the degree of freedom, is the average target intensity, Γ(·) is the gamma function, which is: Γ(x)=∫0 +∞ t x-1 e -t dt(x>0) t is the integration variable.

9. The underwater target echo detection system based on amplitude fluctuation characteristics according to claim 7, characterized in that: Module M3 includes: Module M3.1: further process the echo data of the target detection unit, and calculate the probability density of the echo signal y of the target detection unit under the conditions of H0 and H1 respectively; Under the H0 condition: Under H1 conditions: Among them, f(c) is the probability density function of noise, g(yc) is the probability density of target echo amplitude fluctuation, Module M3.2: Calculate the likelihood ratio function value: λ(y)=p(y|H1) / p(y|H0); Module M4 includes: Module M4.1: Obtain the echo signal z of each reference unit, calculate the corresponding λ(z) value, and calculate the probability density function p(λ(z)|H0) of λ(z) under the condition of H0; Module M4.2: Calculate the threshold μ based on the calculated parameters of the echo signal z, which is specifically determined by the following formula: P fa =∫ μ ∞ p(λ(z)|H0)dλ(z)=α Wherein, α is a preset false alarm probability value, the threshold μ is the lower limit of the integral function, μ ≥ 0 and is adaptive.

10. The underwater target echo detection system based on amplitude fluctuation characteristics according to claim 7, characterized in that: Module M5 includes: Module M5.1: Determine the value of the probability function according to the Neyman-Pearson criterion and classify the likelihood ratio function into Module M5.2: Compare the likelihood ratio function value with the threshold value. When the likelihood ratio function is greater than μ, retain the echo signal corresponding to the target detection unit. When the likelihood ratio function is less than μ, assign a uniform preset value to the corresponding echo signal. Thus, the target detection result is obtained; Module M5.3: The target detection result is formed into a detection echo image, the sliding window position is changed, and the module M2-module M5 are repeatedly triggered until all target detection units are detected, and the final detected echo image is output.

Images