Underwater target detection method based on multi-source data fusion, program, equipment and storage medium

Abstract

The invention belongs to the technical field of signal processing, and particularly relates to an underwater target detection method based on multi-source data fusion, a program, equipment and a storage medium. According to the method, confidence is quantified through a fuzzy set and a membership function, and compared with a hard decision, ground surface channel information is finer through soft decision fusion, so that a 0-1 discretization error of the hard decision is avoided, and the target judgment accuracy is improved; a membership function with adjustable parameters is introduced, and a mapping relation between a non-confidence interval and confidence can be dynamically adjusted according to the characteristics of the sensor to adapt to different noise environments and signal ranges; a uniform quantizer is adopted to replace an optimal quantizer, a complex optimization process is avoided, meanwhile, the calculation complexity is reduced through a log-likelihood ratio fusion rule, and the real-time processing requirement is met; and a confidence interval division mechanism is adopted, so that the sensor outputs a hard decision at high confidence, a soft decision is output in a middle region, and the noise robustness and the information utilization rate are balanced.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, specifically relating to underwater target detection methods, programs, devices, and storage media based on multi-source data fusion. Background Technology

[0002] Although acoustic detection remains the most effective method for underwater applications from a physical perspective, the limitations and potential failures of single-sensor detection in certain situations have made combining multi-source data for underwater target detection an important development trend. Underwater target detection methods based on multi-source data fusion primarily address the task of target detection within the overall target assessment process.

[0003] Multi-source data fusion can comprehensively utilize information acquired by different sensors, effectively avoiding the limitations and uncertainties of a single sensor, thereby improving the overall detection performance of the system. In a target detection system, sensors detect targets in background noise by receiving echo signals from targets within their study range or noise generated by the targets, and then process the received observations to determine the presence of the target. In a multi-sensor detection system with data fusion, each sensor monitors a specific variable and sends its independent hard (or soft) decision to a central processor (fusion center). The central processor fuses the hard (or soft) decisions of the sensors into a single overall (global) decision. Because such systems simultaneously integrate observation information from multiple sensors, they exhibit better performance in terms of reliability and detection capabilities compared to single-sensor systems.

[0004] The combined acoustic-electromagnetic detection platform consists of a multi-beam forward-looking sonar (acoustic detection subsystem) and a loop-source electromagnetic detection subsystem. The loop-source electromagnetic detection subsystem employs an active-source electromagnetic detection method. Since underwater targets are generally constructed of metallic materials to ensure structural strength, electromagnetic methods are highly sensitive to metals, making detection relatively easy. Because underwater target detection involves electromagnetic detection in conductive media, the detection theories of air-to-air radar are not applicable; electromagnetic detection in conductive media is a problem of electromagnetic induction. The loop-source radiation source is a small-volume radiation source. Since its radiation capability depends on the number of turns in the loop coil, multiple turns can be used to enhance the radiation source's capability. Compared to electrical radiation sources, the loop-source is smaller, and the electromagnetic detection subsystem uses the loop-source as its radiation source. Summary of the Invention

[0005] The purpose of this invention is to provide an underwater target detection method, program, device, and storage medium based on multi-source data fusion. This invention integrates observation information from multiple sensors to provide reliability and detection performance. For the acoustic-electromagnetic joint detection platform, a GLRT model based on fluxgate sensor output data is established. A decision matrix is ​​constructed using multi-level quantized soft decision-making, which aims to reduce information loss caused by hard decision fusion compared to centralized fusion and further improve detection performance.

[0006] The underwater target detection method based on multi-source data fusion includes the following steps:

[0007] Deploy a multi-source sensor array, including a sonar array and a fluxgate sensor array;

[0008] The detection probability and false alarm probability of each sonar element in the sonar array are obtained, and the detection threshold of each sonar element is set based on the false alarm probability of each sonar element; the detection probability and false alarm probability of each fluxgate sensor element in the fluxgate sensor array are obtained; the detection threshold of each fluxgate sensor is set based on the false alarm probability of each fluxgate sensor element; and the global detection threshold is set according to the expected global false alarm probability of the underwater target.

[0009] The underwater target detection task is performed within the detection range of the multi-source sensor array. The echo data matrix and beamforming manifold matrix received by each sonar element in the sonar array are obtained. The noise complex Gaussian random distribution parameters and target azimuth matrix of each sonar element are estimated, and then the detection statistics of each sonar element are calculated. The observation values ​​of each fluxgate sensor element in the fluxgate sensor array are obtained, and the detection statistics of each fluxgate sensor element are calculated.

[0010] The detection statistics of each sonar array element and the detection statistics of each fluxgate sensor array element are converted into corresponding confidence levels. Based on the detection statistics and detection thresholds of each sonar array element and each fluxgate sensor array element, the confidence levels of each sonar array element and each fluxgate sensor array element are weighted and summed to obtain the total fused confidence level.

[0011] If the total fusion confidence score is greater than the global detection threshold, it is determined that the multi-source sensor array has detected an underwater target; otherwise, the detection range of the multi-source sensor array is adjusted, and the underwater target detection task is re-executed.

[0012] Furthermore, the sonar array has N1 array elements, and the detection probability P of each sonar array element in the sonar array is obtained. d1i With the false alarm probability P f1i Based on the false alarm probability P of each sonar array element f1i Set the detection threshold T for each sonar element. 1i , i = 1, 2, ..., N1;

[0013] The fluxgate sensor array has N2 array elements, and the detection probability P of each fluxgate sensor element in the array is obtained. d2j With the false alarm probability P f2j Based on the false alarm probability P of each fluxgate sensor array element f2j Set the detection threshold T for each fluxgate sensor. 2j j = 1, 2, ..., N2;

[0014] The global detection threshold T0 is set based on the expected global false alarm probability of the underwater target.

[0015] Furthermore, the echo data matrix R received by each sonar element in the sonar array is obtained. i Beamforming manifold matrix H i Estimate the parameters of the complex Gaussian random distribution of noise. With target azimuth matrix Then, the detection statistics Λ of each sonar array element are calculated. i :

[0016]

[0017] in, k p is the number of snapshots, r is the number of beams; tr[·] represents the trace of the computation matrix; I is the identity matrix.

[0018] Furthermore, the observed values ​​of each fluxgate sensor element in the fluxgate sensor array are obtained. Calculate the detection statistics L for each fluxgate sensor element. j :

[0019]

[0020] Where, x jn is the nth observation of the jth fluxgate sensor element in the fluxgate sensor array, where n = 1, 2, ..., N3, and N3 is the total number of data collected by the fluxgate sensor array.

[0021] Furthermore, the conversion of the detection statistics of each sonar array element and the detection statistics of each fluxgate sensor array element into corresponding confidence levels specifically involves:

[0022] Statistical analysis of the detection statistic Λ for all sonar array elements i Determine the value range [min(Λ)] i ),max(Λ i Within this value range, set Q1-1 dividing points t. aqThe range of values ​​is divided into Q1 equal parts, with each part having a step size of ε1.

[0023] The detection statistics Λ of each sonar array element i Convert to the corresponding confidence level δ Λi :

[0024]

[0025] in, t aq =qε1, q = 1, 2, ..., Q1-1;

[0026] Statistical analysis of the detection statistics L for all fluxgate sensor array elements j Determine the value interval [min(L)] j ),max(L j Within this value range, set Q2-1 dividing points t. bp The range of values ​​is divided into Q2 equal parts, with each part having a step size of ε2.

[0027] The detection statistics L of each fluxgate sensor array element j Convert to the corresponding confidence level δ Lj :

[0028]

[0029] in, t bp =pε2, p = 1, 2, ..., Q2-1.

[0030] Furthermore, based on the detection statistics Λ of each sonar array element i With detection threshold T 1i Detection statistics L of each fluxgate sensor array element j With detection threshold T 2j The confidence level δ for each sonar array element Λi The confidence level δ of each fluxgate sensor array element Lj We perform a weighted summation to obtain the total fusion confidence δ. Z ;

[0031]

[0032] Among them, w Λi w represents the weight of the i-th sonar element. Lj Let be the weight of the j-th fluxgate sensor element.

[0033] Furthermore, the weight w of the i-th sonar element ΛiThe weight w of the j-th fluxgate sensor element Lj for:

[0034]

[0035] A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described underwater target detection method based on multi-source data fusion.

[0036] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described underwater target detection method based on multi-source data fusion.

[0037] A computer program product includes computer instructions that, when executed by a processor, implement the steps of the above-described underwater target detection method based on multi-source data fusion.

[0038] The beneficial effects of this invention are as follows:

[0039] This invention quantizes confidence levels using fuzzy sets and membership functions. Compared to hard decision-making, soft decision fusion provides a more refined representation of channel information, avoiding the "0-1" discretization error of hard decision-making and improving target decision accuracy. The introduction of adjustable membership functions allows for dynamic adjustment of the mapping relationship between the confidence interval and confidence level based on sensor characteristics, adapting to different noise environments and signal ranges. A uniform quantizer replaces the optimal quantizer, avoiding complex optimization processes, while a log-likelihood ratio fusion rule reduces computational complexity, meeting real-time processing requirements. The confidence interval division mechanism enables the sensor to output hard decisions at high confidence levels and soft decisions in the intermediate region, balancing noise robustness and information utilization. Attached Figure Description

[0040] Figure 1 This is a connection diagram of the electromagnetic detection subsystem of the return source.

[0041] Figure 2 This is a layout diagram of the return-source electromagnetic detection system.

[0042] Figure 3 Image of forward-looking sonar and the target being measured.

[0043] Figure 4 This is a layout diagram of the acoustic detection system.

[0044] Figure 5 This is the overall flowchart for joint testing.

[0045] Figure 6 This is a graph showing the relationship between membership functions and likelihood ratios.

[0046] Figure 7This is the transfer characteristic diagram of an octet quantizer.

[0047] Figure 8 This is a probability distribution curve of the observed data when the power-to-noise ratio is -1dB.

[0048] Figure 9 The ROC curve is shown when the power-to-noise ratio is -1dB and the number of sensors is 3.

[0049] Figure 10 The ROC curve is shown when the power-to-noise ratio is -1dB and the number of sensors is 4.

[0050] Figure 11 The ROC curve is shown when the power-to-noise ratio is -1dB and the number of sensors is 5.

[0051] Figure 12 This is a probability distribution curve of the observed data when the power-to-noise ratio is -3dB.

[0052] Figure 13 The ROC curve is shown when the power-to-noise ratio is -3dB and the number of sensors is 4.

[0053] Figure 14 This is a probability distribution curve of the observed data when the power-to-noise ratio is 1dB.

[0054] Figure 15 The ROC curve is shown when the power-to-noise ratio is 1dB and the number of sensors is 4.

[0055] Figure 16 The graph shows the change of PD with power-to-noise ratio when PFA = 0.1.

[0056] Figure 17 The diagram shows the results of the forward-looking sonar detection domain division under two different conditions.

[0057] Figure 18 This is a graph showing the ROC curves for different detection methods in the field test.

[0058] Figure 19 This is a graph showing the performance of the sonar subsystem as a function of the threshold during field testing.

[0059] Figure 20 This is a graph showing the performance of the electromagnetic subsystem as a function of the threshold during field testing. Detailed Implementation

[0060] The present invention will now be further described with reference to the accompanying drawings.

[0061] This invention studies target detection methods in underwater target detection scenarios. It mainly establishes detection models by separately using multibeam forward-looking sonar echo data and fluxgate sensor output data in the electromagnetic detection subsystem, and designs a single-sensor detection method. Based on the data characteristics of the two sensors and the detection results, multi-sensor data fusion technology is used for joint detection. A multi-level quantitative soft decision fusion method is designed based on the acoustic-electromagnetic joint detection platform.

[0062] I. Establishing a GLRT model based on multibeam sonar echo data;

[0063] The detection problem of multibeam forward-looking sonar can be regarded as a binary hypothetical detection problem, where H0 represents the case without a target and H1 represents the case with a target. Assume that each sonar array has N elements, and the k-th snapshot signal received by an element is denoted as H0. k p This represents the number of snapshots contained in a ping. The noise follows a complex Gaussian random distribution. And λ is unknown. Beamforming manifold matrix: Target azimuth beam vector: r represents the number of beams.

[0064] Based on the above, the binary hypothesis testing model for the multibeam forward-looking sonar at time k is as follows:

[0065]

[0066] Based on the detection and improvement model of 1-ping array data, for a total of k within 1 ping p For the array signal of a snapshot, the echo data received by the array is denoted as... The target azimuth matrix is ​​denoted as The GLRT method for multi-beam forward-looking sonar detection is derived as follows:

[0067]

[0068] Among them, f j (R;·) is R in H j The probability density function for j = 0, 1, where T is based on the preset false alarm probability (P). fa The corresponding detection threshold.

[0069] For this expression, we first solve the corresponding optimization problem under the condition H0. The likelihood function f0(R; λ) can be written as:

[0070]

[0071] Where tr(·) denotes the trace of the matrix, which is solved using the maximum likelihood estimation method. This can be expressed as:

[0072]

[0073] After solving the optimization problem, we obtain The expression, then Substituting the known parameters into the likelihood function yields... The expression:

[0074]

[0075] For solving the corresponding optimization problem under case H1, the likelihood function f1(R; P, λ) can be written as:

[0076]

[0077] Estimator of λ It can be represented as:

[0078]

[0079] Will Substituting the known parameters into the likelihood function yields... Estimator of P It can be represented as:

[0080]

[0081] Calculating (8) is equivalent to calculating the following problem:

[0082]

[0083] For the pseudo-inverse of H, Substituting the known parameters into the likelihood function yields The expression, where:

[0084]

[0085] therefore:

[0086]

[0087] Finally, we obtain the generalized likelihood ratio expression as the detection statistic:

[0088]

[0089] When discussing only the GLRT problem of a single multibeam sonar, for computational convenience, it can be further simplified to the form of the detection statistic:

[0090]

[0091] II. Establishing a GLRT model based on data from a loop source electromagnetic detection system;

[0092] A GLRT model was established based on the observation data of the fluxgate sensor in the loop source electromagnetic detection system. However, unlike previous models, this model was further improved to a high / low level detection model where the amplitude of the voltage level and the variance of the noise are unknown, taking into account the actual situation. The model established for the detection problem of the loop source electromagnetic detection system is as follows:

[0093]

[0094] Where x[n] represents the observed value of the fluxgate sensor, A is unknown, -∞ < A < +∞, and w[n] is white Gaussian noise (WGN) with variance σ 2 It is unknown, N is the total number of data collected;

[0095] The GLRT method for the fluxgate sensor model is derived as follows:

[0096]

[0097] Among them, f j (x;·) is x in H j The probability density function for j = 0, 1, where γ is based on the preset false alarm probability (P). fa The corresponding detection threshold.

[0098] First, we solve the optimization problem under case H0. According to formula (14), the likelihood function f0(x; σ) 2 This can be represented as:

[0099]

[0100] Therefore σ 2 estimator It can be expressed as:

[0101]

[0102] By solving the optimization problem, we can obtain... Will Substituting the known variables into the likelihood function, we obtain... The expression:

[0103]

[0104] To solve the optimization problem in case H1, according to formula (14), the likelihood function f1(x; A, σ) 2 This can be represented as:

[0105]

[0106] Therefore, the estimator of A It can be expressed as:

[0107]

[0108] in:

[0109]

[0110] By solving the optimization problem, we obtained... Next Substituting the known parameters into the likelihood function yields... σ 2 estimator It can be expressed as:

[0111]

[0112] By solving the optimization problem, we obtained... Next Substituting the known parameters into the likelihood function yields... The expression:

[0113]

[0114] To differentiate between the two hypotheses H0 and H1, variance estimates are obtained through estimation. Next, we will assume σ under H0. 2 The estimate is denoted as Under the assumption of H1, σ 2 The estimate is denoted as

[0115] Substituting formulas (18) and (23) into the generalized likelihood ratio expression, we can obtain the detection statistic as follows:

[0116]

[0117] After taking the logarithm:

[0118]

[0119] in:

[0120]

[0121] Third, based on the sonar GLRT model and the loop source electromagnetic detection system GLRT model, a multi-level quantitative soft decision fusion method is constructed to complete a high-performance fusion detection method;

[0122] Soft decision δ i It is through the use of L i A fuzzy set A is defined i A was generated from this. i It is an ordered pair:

[0123]

[0124] in, Known as L i In fuzzy set A i The membership degree or membership function in the likelihood ratio space L maps the likelihood ratio space L to the interval [0,1]. If the membership function contains only two values, 0 and 1, then A i It is a non-fuzzy set. This will be the same as in the hard decision case. If sensor i tends to assume H1, The value will be greater than 0.5 (the corresponding hard decision is u). i =1), if the sensor tends to assume H0, The value will be less than 0.5 (the corresponding hard decision is u). i =0). The relationship between soft decision-making and the corresponding sensor hard decision-making and the likelihood ratio test is expressed as:

[0125]

[0126] For a given sensor i, use a scalar quantizer Q i The value δ of its membership function i Quantization is performed. The scalar quantizer Q... i It is a mapping that uses a q-level quantizer to quantize the input variable δ i (Within the interval [0,1]) is mapped to the output variable δ ij (Also within the interval [0,1]), where j = 1, 2, ..., q. Each endpoint of the quantization interval j has a corresponding threshold, denoted as t. i(j-1) and t ij And there is a corresponding soft decision value δ ij (Quantizer output). The quantization level (q) depends on the information rate R of each channel. i The constraints are set to i = 1, 2, ..., n. If the number of bits per quantization level is m... i (q i =2 mi If the communication constraints are satisfied, then:

[0127] 0 < 2 mi ≤R i i = 1, 2, ..., n (29)

[0128] This invention only considers using a uniform quantizer. Therefore, the membership values ​​within the interval [0,1] are divided with a uniform step size of ε (ε = 1 / q). If the value of the membership function falls within the j-th quantization interval, then the quantized value is taken as the midpoint of that interval. This is used to generate the quantization membership level δ. ij The transfer characteristics of a uniform quantizer are defined as follows:

[0129]

[0130] The soft decision rule for sensor i can then be described as follows:

[0131]

[0132] Having obtained the soft decision-making rules for each sensor, we will now discuss the soft decision fusion problem. Let Ω i Let i be the local decision space of sensor i. For u i The local hard decision space when =1, For u i The local hard decision space when = 0. Then we have:

[0133]

[0134] Similar to the definition of local hard decision, we will now consider the soft decision space. For sensor i, let... For δ i The local soft decision space when = 0, For δ i =δ ij The local soft decision space when <0.5, j=1,2,...,q / 2 For δ i =δ ij >0.5, j=(q / 2+1),(q / 2+2),...,q, and For δ i The local soft decision space when = 1. The soft decision values ​​are as follows:

[0135]

[0136] In this situation, the local soft decision-making space and It can be represented as:

[0137]

[0138] The fusion center uses all sensor soft decision values ​​δ = (δ1, δ2, ..., δ) output from each sensor. nTo perform the NP test, that is, to construct a likelihood ratio test (assuming the observations are independent):

[0139]

[0140] Assume n Ho One sensor determines 0, and n1 sensors determine δ. i1 n2 sensors determine δ i2 And so on, n q The sensor determined that δ iq , The sensor determines 1, and:

[0141]

[0142] Formula (36) can be written as:

[0143]

[0144] By taking the logarithm, we can obtain:

[0145]

[0146] The rules for overall decision-making at the integration center are as follows:

[0147]

[0148] Where t0 is the threshold of the fusion center, which is determined based on the expected global false alarm probability (GFAP), and the coefficient {a} ij}, i=1,2,...,n,j=1,2,...,q is based on the membership degree {δ} of sensor i i The confidence level is determined by the following: i = 1, 2, ..., n (or equivalently, by the confidence level).

[0149] Based on the decision rules of different decision fusion methods, Monte Carlo simulation obtains the decision results of the corresponding rules by simulating a large number of decision processes. Then, it estimates the false alarm probability and detection probability, which can reflect the detection performance, by statistically analyzing the decision results, so as to compare and analyze different fusion methods.

[0150] Based on the above theoretical derivation, the underwater target detection method based on multi-source data fusion in this invention includes the following steps:

[0151] Step 1: Deploy a multi-source sensor array, including a sonar array and a fluxgate sensor array. The sonar array has N1 array elements, and the fluxgate sensor array has N2 array elements.

[0152] Obtain the detection probability P of each sonar element in the sonar array. d1iWith the false alarm probability P f1i , i = 1, 2, ..., N1; Obtain the detection probability P of each fluxgate sensor element in the fluxgate sensor array. d2j With the false alarm probability P f2j j = 1, 2, ..., N2;

[0153] Based on the false alarm probability P of each sonar element f1i Set the detection threshold T for each sonar element. 1i Based on the false alarm probability P of each fluxgate sensor array element f2j Set the detection threshold T for each fluxgate sensor. 2j ;

[0154] Set the global detection threshold T0 based on the expected global false alarm probability of the underwater target;

[0155] Step 2: The multi-source sensor array performs underwater target detection, acquiring the echo data matrix received by each sonar element in the sonar array. Beamforming manifold matrix H i ; Obtain the observation values ​​of each fluxgate sensor element in the fluxgate sensor array.

[0156] in, k p γ represents the number of snapshots, r represents the number of beams; ik Let be the k-th snapshot signal received by the i-th sonar element in the sonar array. x jn is the nth observation of the jth fluxgate sensor element in the fluxgate sensor array, where n = 1, 2, ..., N3, and N3 is the total number of data collected by the fluxgate sensor array.

[0157] Step 3: For each sonar element in the sonar array, estimate the parameters of the complex Gaussian random distribution of noise. With target azimuth matrix Then, the detection statistic Λ of the sonar array element is calculated. i ;

[0158]

[0159]

[0160] Where tr[·] denotes the trace of the computed matrix; I is the identity matrix;

[0161] Step 4: For each fluxgate sensor element in the fluxgate sensor array, based on the collected N3 sets of observation values ​​x jn Calculate the detection statistic L j ;

[0162]

[0163] in,

[0164] Step 5: Calculate the detection statistics Λ for all sonar array elements. i Determine the value range [min(Λ)] i ),max(Λ i Within this value range, set Q1-1 dividing points t. aq The range of values ​​is divided into Q1 equal parts, with each part having a step size of ε1.

[0165]

[0166] Set confidence level δ 1q , The detection statistics Λ of each sonar array element i Convert to the corresponding confidence level δ Λi ;

[0167]

[0168] Step 6: Calculate the detection statistics L for all fluxgate sensor array elements. j Determine the value interval [min(L)] j ),max(L j Within this value range, set Q2-1 dividing points t. bp The range of values ​​is divided into Q2 equal parts, with each part having a step size of ε2.

[0169]

[0170] Set confidence level δ 2p , The detection statistics L of each fluxgate sensor array element j Convert to the corresponding confidence level δ Lj ;

[0171]

[0172] Step 7: Based on the detection statistics Λ of each sonar element i With detection threshold T 1i Detection statistics L of each fluxgate sensor array element j With detection threshold T 2j The confidence level δ for each sonar array element Λi The confidence level δ of each fluxgate sensor array element Lj We perform a weighted summation to obtain the total fusion confidence δ. Z ;

[0173]

[0174] Among them, w Λi w represents the weight of the i-th sonar element. Lj Let be the weight of the j-th fluxgate sensor element;

[0175]

[0176] If the total fusion confidence level δ Z Greater than the global detection threshold T0, i.e., δ Z If the value is greater than T0, it is determined that the multi-source sensor array has detected an underwater target; otherwise, the detection range of the multi-source sensor array is adjusted, and the underwater target detection task is re-executed.

[0177] This invention establishes a GLRT model based on fluxgate sensor output data for an acoustic-electromagnetic joint detection platform. Each sensor sends multi-level quantized soft decisions to the fusion center. Sensor soft decisions are derived, with the soft decision value proportional to the confidence level of the judgment hypothesis, and its value depending on the difference between the likelihood ratio and the threshold. Soft decisions are generated by defining fuzzy sets, and the actual value depends on the no-confidence interval and the expected signal range under the assumed conditions. For a given sensor, a scalar quantizer (considering a uniform quantizer) is used to quantize its membership function value, obtaining the soft decision judgment rule for each sensor. The fusion center then performs an NP test using the soft decision values ​​of each sensor to construct a likelihood ratio test. In the overall decision rule, the threshold is determined based on the expected global false alarm probability, and the coefficients are determined based on the sensor membership degree. The use of multi-level quantized soft decisions to construct a decision matrix aims to reduce information loss caused by hard decision fusion compared to centralized fusion, further improving detection performance.

[0178] Example 1:

[0179] To ensure that the estimated indicators based on statistical results closely approximate actual performance, this embodiment uses a relatively large number of iterations (10,000 iterations, where each iteration is considered a complete process from individual sensor decisions to data fusion in both H0 and H1 scenarios) to generate random samples. Each sensor makes L decisions in one iteration. To output an ROC curve, simulations are performed using different preset false alarm probabilities based on the NP criterion. This is achieved by changing the preset false alarm probability P. FA (P in this experiment) FA Sixty-five values ​​were uniformly selected within the range [0, 0.99], representing 65 different preset false alarm probabilities, and simulations were performed 10,000 times per iteration. Monte Carlo simulations were conducted according to the simulation process described above, and ROC curves were obtained to evaluate the performance of various fusion methods. Figure 8 , Figure 9 , Figure 10 and Figure 11 The relevant information is given when the power-to-noise ratio of the Monte Carlo simulation is -1dB. Figure 8 The probability distribution curves of the observed data under the assumptions and under the assumptions are shown when the power-to-noise ratio is -1dB.

[0180] Figure 9 , Figure 10 , Figure 11 ROC curves for various methods are presented for three scenarios: an energy-to-noise ratio of -1dB and 3, 4, and 5 sensors. Analysis... Figure 9 , Figure 10 , Figure 11 The ROC curves show that the combined detection of several sensors demonstrates better performance compared to single-sensor detection, reflecting the improvement in detection performance brought about by multi-sensor data fusion. Centralized data fusion methods show the best performance, followed by soft-decision fusion methods. This demonstrates that in data fusion, even at the cost of communication bandwidth, utilizing more methods to extract sensor observations often yields better performance.

[0181] By comparison Figure 10 , Figure 13 , Figure 15 It can be observed that with the increase of the power-to-noise ratio, the performance of various data fusion methods is significantly improved, and the performance gap between other data fusion methods and centralized fusion methods is further narrowed. Compared with centralized fusion methods, soft decision fusion occupies less communication bandwidth and has less computational load, effectively improving detection performance. A GLRT detector and a multi-level quantization soft decision fusion method were designed for joint detection of a multi-beam forward-looking sonar and a loop-source electromagnetic detection subsystem. An acoustic-electromagnetic joint detection experiment was designed and completed. The multi-sensor detection system used in this joint detection experiment consists of a dual-frequency multi-beam forward-looking sonar and a loop-source electromagnetic detection subsystem. The experimental environment selected for this experiment was an open freshwater area located in Songhua Lake, Fengman District, Jilin City, Jilin Province. The lake test simulated common application scenarios of acoustic-electromagnetic joint detection, and the acoustic-magnetic joint detection method was verified.

[0182] exist Figure 18 It can be observed that, in practical underwater target detection tasks, when there are certain differences in the detection performance of two sensors, the multi-sensor joint detection method can largely compensate for the performance degradation caused by the relatively poor detection performance of the sensor, leading to better overall decision-making. The detection performance of the multi-sensor detection system is better and more stable than that of any single subsystem within the system. Furthermore, the multi-level quantization soft decision fusion method proposed in this invention exhibits good detection performance, especially when the false alarm probability is low, the improvement in detection probability is even more significant. Figure 19 and Figure 20 This provides a reference for selecting an appropriate threshold when applying the method. When using the same method to construct the detection statistic, the selection of the detection threshold should be as close as possible to the appropriate threshold. Figure 19 and Figure 20 Finding a balance point is crucial; the threshold should be chosen to minimize the false alarm probability while maximizing the detection probability. The two figures show that the sonar subsystem is more likely to select a satisfactory detection threshold compared to the electromagnetic detection subsystem. This also corresponds to... Figure 19 In the ROC curve, the detection performance of the sonar subsystem is better than that of the electromagnetic detection subsystem.

[0183] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An underwater target detection method based on multi-source data fusion, characterized in that: Deploy a multi-source sensor array, including a sonar array and a fluxgate sensor array; The detection probability and false alarm probability of each sonar element in the sonar array are obtained, and the detection threshold of each sonar element is set based on the false alarm probability of each sonar element; the detection probability and false alarm probability of each fluxgate sensor element in the fluxgate sensor array are obtained; the detection threshold of each fluxgate sensor is set based on the false alarm probability of each fluxgate sensor element; and the global detection threshold is set according to the expected global false alarm probability of the underwater target. The underwater target detection task is performed within the detection range of the multi-source sensor array. The echo data matrix and beamforming manifold matrix received by each sonar element in the sonar array are obtained. The noise complex Gaussian random distribution parameters and target azimuth matrix of each sonar element are estimated, and then the detection statistics of each sonar element are calculated. The observation values ​​of each fluxgate sensor element in the fluxgate sensor array are obtained, and the detection statistics of each fluxgate sensor element are calculated. The detection statistics of each sonar array element and the detection statistics of each fluxgate sensor array element are converted into corresponding confidence levels. Based on the detection statistics and detection thresholds of each sonar array element and each fluxgate sensor array element, the confidence levels of each sonar array element and each fluxgate sensor array element are weighted and summed to obtain the total fused confidence level. If the total fusion confidence score is greater than the global detection threshold, it is determined that the multi-source sensor array has detected an underwater target; Otherwise, adjust the detection range of the multi-source sensor array and re-execute the underwater target detection task.

2. The underwater target detection method based on multi-source data fusion according to claim 1, characterized in that: The sonar array has N1 array elements, and the detection probability P of each sonar element in the sonar array is obtained. d1i With the false alarm probability P f1i Based on the false alarm probability P of each sonar array element f1i Set the detection threshold T for each sonar element. 1i , i = 1, 2, ..., N1; The fluxgate sensor array has N2 array elements, and the detection probability P of each fluxgate sensor element in the array is obtained. d2j With the false alarm probability P f2j Based on the false alarm probability P of each fluxgate sensor array element f2j Set the detection threshold T for each fluxgate sensor. 2j j = 1, 2, ..., N2; The global detection threshold T0 is set based on the expected global false alarm probability of the underwater target.

3. The underwater target detection method based on multi-source data fusion according to claim 2, characterized in that: Obtain the echo data matrix R received by each sonar element in the sonar array. i Beamforming manifold matrix H i Estimate the parameters of the complex Gaussian random distribution of noise. With target azimuth matrix Then, the detection statistics Λ of each sonar array element are calculated. i : in, k p is the number of snapshots, r is the number of beams; tr[·] represents the trace of the computation matrix; I is the identity matrix.

4. The underwater target detection method based on multi-source data fusion according to claim 2, characterized in that: Acquire the observation values ​​of each fluxgate sensor element in the fluxgate sensor array. Calculate the detection statistics L for each fluxgate sensor element. j : Where, x jn is the nth observation of the jth fluxgate sensor element in the fluxgate sensor array, where n = 1, 2, ..., N3, and N3 is the total number of data collected by the fluxgate sensor array.

5. The underwater target detection method based on multi-source data fusion according to claim 2, characterized in that: The conversion of the detection statistics of each sonar array element and the detection statistics of each fluxgate sensor array element into corresponding confidence levels is specifically as follows: Statistical analysis of the detection statistic Λ for all sonar array elements i Determine the value range [min(Λ)] i ),max(Λ i Within this value range, set Q1-1 dividing points t. aq The range of values ​​is divided into Q1 equal parts, with each part having a step size of ε1. The detection statistics Λ of each sonar array element i Convert to the corresponding confidence level δ Λi : in, t aq =qε1, q = 1, 2, ..., Q1-1; Statistical analysis of the detection statistics L for all fluxgate sensor array elements j Determine the value range [min(L)] j ),max(L j Within this value range, set Q2-1 dividing points t. bp The range of values ​​is divided into Q2 equal parts, with each part having a step size of ε2. The detection statistics L of each fluxgate sensor array element j Convert to the corresponding confidence level δ Lj : in, t bp =pε2, p = 1, 2, ..., Q2-1.

6. The underwater target detection method based on multi-source data fusion according to claim 5, characterized in that: Based on the detection statistics Λ of each sonar array element i With detection threshold T 1i Detection statistics L of each fluxgate sensor array element j With detection threshold T 2j The confidence level δ for each sonar array element Λi The confidence level δ of each fluxgate sensor array element Lj We perform a weighted summation to obtain the total fusion confidence δ. Z ; Among them, w Λi w represents the weight of the i-th sonar element. Lj Let be the weight of the j-th fluxgate sensor element.

7. The underwater target detection method based on multi-source data fusion according to claim 5, characterized in that: The weight w of the i-th sonar element Λi The weight w of the j-th fluxgate sensor element Lj for:

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

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

10. A computer program product comprising computer instructions, characterized in that: When executed by a processor, the computer instructions implement the steps of the method according to any one of claims 1 to 7.

Images