Intelligent underwater acoustic signal processing system and method based on multi-sensor data fusion

Abstract

The present invention discloses an intelligent underwater acoustic signal processing system and method based on multi-sensor data fusion, belonging to the field of underwater signal processing. Acoustic signals, linear acceleration, angular velocity and other information of underwater targets are collected by multiple sensors and continuous analog signals are output, and then transmitted to a signal conditioning module to dynamically adjust the gain according to the target strength, and the amplified differential signal is output to an analog-to-digital conversion module. High-precision digitization is performed through a 24-bit Δ-Σ ADC and a digital signal is output to a main control unit, which performs multi-sensor data fusion and target azimuth calculation; finally, the main control unit uploads the underwater acoustic signal and device attitude information data to a host computer through an RS485 bus for human-computer interaction. Based on multi-sensor data fusion, this solution can achieve high-precision target detection and positioning in a dynamic underwater environment and has wide popularization and application value.

Description

Technical Field

[0001] The present invention belongs to the field of underwater signal processing, and particularly relates to an intelligent underwater acoustic signal processing system and method based on multi-sensor data fusion. Background Art

[0002] With the continuous advancement of fields such as marine resource development, subsea engineering monitoring, and underwater security prevention, the requirements for the accuracy and real-time performance of underwater target detection and positioning are constantly increasing. Traditional underwater detection systems mainly rely on sonar technology, collect underwater acoustic signals through a hydrophone array, and then calculate the target position using parameters such as phase and amplitude differences.

[0003] However, in practical applications, traditional systems have many deficiencies:

[0004] On the one hand, most systems adopt a single sensor or a multi-sensor scheme with a simple combination. Due to the differences in the acquisition timing, sensitivity, and response speed of each sensor, the data fusion effect is not good, errors are easily introduced, and the positioning accuracy is affected. On the other hand, the underwater environment is complex and variable, and factors such as temperature, pressure, flow velocity, and ocean clutter will exacerbate the background noise. Traditional systems lack effective dynamic calibration and error compensation mechanisms and are difficult to accurately extract weak target signals.

[0005] In addition, underwater vibration or motion signals themselves are very weak and must be processed after multiple stages of amplification, filtering, and high-precision analog-to-digital conversion. Existing signal conditioning schemes often have delays in anti-interference, real-time acquisition, and digitization processes and cannot meet the requirements of fast response. Moreover, underwater data transmission is vulnerable to communication link interference, with a high bit error rate and obvious delay during the transmission process, further restricting the real-time monitoring and remote control capabilities of the system. Summary of the Invention

[0006] Aiming at the problems existing in the prior art, such as insufficient multi-sensor collaboration accuracy, limited adaptability to dynamic environments, and lack of real-time emergency processing capabilities, the present invention provides an intelligent underwater acoustic signal processing system and method based on multi-sensor data fusion. Through the collaborative processing of multiple sensors, such as hydrophones, underwater acceleration sensors, attitude sensors, and temperature sensors, and the optimization of intelligent algorithms, high-precision detection in a dynamic underwater environment is achieved.

[0007] The present invention is implemented by the following technical solutions: An intelligent underwater acoustic signal processing system based on multi-sensor data fusion includes a main control unit, a data processing module, and a multi-sensor module connected to the main control unit. The multi-sensor module includes a hydrophone, an underwater acceleration sensor, an attitude sensor, and a temperature sensor, which respectively collect underwater acoustic signals, linear acceleration signals, angular velocity signals, and temperature signals;

[0008] The data processing module includes a signal conditioning module and an analog-to-digital conversion module. The analog signals collected by the sensor module are amplified and conditioned by the signal conditioning module, and the amplified and conditioned signals are sent to the analog-to-digital conversion module for digitization;

[0009] The main control unit includes a data fusion module, a bearing calculation module, and a calibration module. The data fusion module receives the output information of the data processing module for data fusion, and inputs the data fusion result into the bearing calculation module. Combining with the Eigen-AMVDR algorithm, beamforming and target direction estimation are performed, and finally the target azimuth angle and motion state parameters are output, and the calibration module is combined to ensure the output accuracy;

[0010] The main control unit is connected to the upper computer through the communication module to achieve data transmission and remote monitoring.

[0011] Further, when the data fusion module performs fusion processing, it includes inertial data fusion and underwater acoustic signal fusion. Specifically:

[0012] Inertial data fusion: First, the linear acceleration and angular velocity signals are fused, and the complementary filtering method is used to calculate the current attitude angle; based on the current attitude angle, a rotation matrix is constructed to perform attitude compensation on the linear acceleration signal; combined with dynamic weight allocation for weighted fusion, and the multi-channel weighted average value is calculated as the fusion result; finally, the corrected linear acceleration data is synthesized;

[0013] Underwater acoustic signal fusion: Using the rotation matrix constructed by attitude compensation, the receiving direction is corrected in real time. The corrected underwater acoustic signals are synchronously encapsulated with the data after inertial data fusion under the same time stamp and attitude reference, and then enter the bearing calculation module.

[0014] Further, the bearing calculation module constructs a covariance matrix for the received signals, performs eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors; the eigenvectors with large eigenvalues correspond to the signal subspace, while the eigenvectors with small eigenvalues correspond to the noise subspace. The received signals are projected onto the signal subspace and the noise subspace using the eigenvectors to separate the signal and noise components;

[0015] In the signal subspace, the Eigen-AMVDR algorithm is applied to calculate the optimal weight vector W, so that the gain in the target direction is maximized while suppressing interference and noise in other directions. Specifically:

[0016] (1) Obtain the fused linear acceleration, angular velocity, and underwater acoustic signal data from the data fusion module, construct sample data according to the predefined time window, and calculate the covariance matrix R using the sampled data i , and then for the covariance matrix R iPerform eigenvalue decomposition to extract the main components of the signal, and obtain the distribution of the signal subspace and the noise subspace;

[0017] (2)Fuse the second-order statistics between multiple channels, and design a weighted fusion power spectrum based on this:

[0018] ;

[0019] Among them, represents the power response function of the i-th channel in the direction , and the weight is determined by the signal energy and its statistical characteristics of each channel. By weighting and fusing the information of multiple channels, the comprehensive beam pointing response is calculated;

[0020] (3)Traverse within the set angle range, calculate the above power spectrum for each angle, and the result can be plotted to form a beam pattern. Weight and fuse the information in the aforementioned signal subspace with the weighted fusion power spectrum;

[0021] Specifically: Define an enhanced beam function:

[0022] ;

[0023] Among them, is the steering vector in the desired direction , is the conjugate transpose of ;

[0024] (4)The actually obtained is regarded as the convolution result of the true target azimuth distribution and a point spread function PSF, that is:

[0025] ;

[0026] Use the deconvolution algorithm to perform iterative processing on to solve the target distribution , and the iterative update formula is:

[0027] ;

[0028] Among them, is initialized by , is the flipped version of ;

[0029] The final target direction estimate is the output result after the convergence of the iterative process:

[0030] ;

[0031] Where K is the number of steps when the iteration termination condition is satisfied, which is updated through iteration, and finally makes the estimated target azimuth Form a clear peak in the correct direction, so as to accurately determine the target azimuth.

[0032] Furthermore, the calibration module is used to achieve real-time dynamic compensation:

[0033] During the operation of the system, the output data of each sensor is continuously collected and compared in real time. The sliding window technology is used to analyze the linear acceleration data collected each time, and its mean and variance are calculated; if the data of multiple consecutive windows exceed the set tolerance range, it is regarded as a deviation; when a deviation is detected, based on the calibration module, a low-pass filter is applied to filter out the sudden noise, and new bias and gain values are recalculated, and the new bias and gain value parameters are dynamically loaded into the calibration module for calibration.

[0034] Furthermore, the signal conditioning module is designed with a differential amplifier circuit. The differential amplifier circuit uses the differential amplifier OPA1632 to construct a variable gain differential amplifier network, including a series-connected first OPA module and a second OPA module. The positive input terminal of the first OPA module is connected to the resistor R1, the negative input terminal is connected to the resistor R2, and the resistor R3 and the capacitor C1 are connected in parallel between the positive input terminal and the negative output terminal of the first OPA module, and the resistor R4 and the capacitor C2 are connected in parallel between the negative input terminal and the positive output terminal. Similarly, the positive input terminal of the second OPA module is connected to the resistor R5, the negative input terminal is connected to the resistor R6, the resistor R7 and the capacitor C3 are connected in parallel between the positive input terminal and the negative output terminal of the second OPA module, the resistor R8 and the capacitor C4 are connected in parallel between the negative input terminal and the positive output terminal of the second OPA module, the negative output terminal of the second OPA module is connected to the capacitor C5, the positive output terminal is connected to the capacitor C6, and the resistor R9 and R10 are respectively connected to the ground corresponding to the negative output terminal and the positive output terminal of the second OPA module.

[0035] Furthermore, after the data processed by the data processing module is transmitted to the main control unit, data preprocessing is first performed, and the preprocessed data is then fused by the data fusion module. When performing data preprocessing, it specifically includes:

[0036] 1) Space-time alignment: By uniformly marking the sampling data of different sensors with time stamps, the time synchronization of multi-source data is realized;

[0037] 2) Static reference calibration: Collect data in a stationary state during the system startup phase to construct a linear calibration model;

[0038] 3) Temperature compensation: Based on the real-time collected temperature data and combined with the temperature drift models of each sensor, the measured values of underwater acoustic signals, angular velocity, and linear acceleration are corrected.

[0039] The present invention further provides an intelligent underwater acoustic signal processing method based on multi-sensor data fusion, including the following steps:

[0040] Step A, data acquisition and preprocessing: The multi-sensor module collects underwater-related signals. The analog signals collected by the sensor module are amplified and conditioned by the signal conditioning module, and the amplified and conditioned signals are sent to the analog-to-digital conversion module for digitization and then uniformly input into the main control unit for signal preprocessing;

[0041] When the main control unit performs signal preprocessing, it includes the following steps:

[0042] 1) Space-time alignment: By uniformly marking the sampling data of different sensors with time stamps, the time synchronization of multi-source data is achieved;

[0043] 2) Static reference calibration: Collect data in a stationary state during the system startup phase to construct a linear calibration model;

[0044] 3) Temperature compensation: Based on the real-time collected temperature data and combined with the temperature drift models of each sensor, the measured values of underwater acoustic signals, angular velocity, and linear acceleration are corrected to reduce the zero-point offset and gain error caused by temperature changes;

[0045] Step B, data fusion and filtering: The data preprocessed by the main control unit undergoes multi-source information fusion through the data fusion module, performs attitude compensation on the linear acceleration signal, and removes noise through filtering;

[0046] Step C, position and direction calculation: In combination with the azimuth calculation module, using the Eigen-AMVDR algorithm, by calculating the covariance matrix of the signal, extracting the eigenvector of the signal, and then estimating the direction and azimuth angle of the target;

[0047] Step D, error calibration and compensation: During the operation of the system, continuously collect the output data of each sensor and perform real-time comparison. Use the sliding window technique to analyze the linear acceleration data collected each time, and calculate its mean and variance; if the data of multiple consecutive windows exceeds the set tolerance range, it is regarded as a deviation; when a deviation is detected, based on the calibration module, apply a low-pass filter to filter out sudden noise, recalculate the new bias and gain values, and dynamically load the new bias and gain value parameters into the calibration module for calibration;

[0048] Step E, data monitoring and display: The finally processed data is displayed and monitored through the upper computer.

[0049] Further, in step C, when performing azimuth calculation, the principle is as follows:

[0050] The azimuth calculation module constructs a covariance matrix for the received signals, performs eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors; the eigenvectors corresponding to large eigenvalues correspond to the signal subspace, while the eigenvectors corresponding to small eigenvalues correspond to the noise subspace. The received signals are projected onto the signal subspace and the noise subspace using the eigenvectors to separate the signal and noise components.

[0051] In the signal subspace, the Eigen-AMVDR algorithm is applied to calculate the optimal weight vector W, so as to maximize the gain in the target direction and suppress interference and noise in other directions. Specifically:

[0052] (1) Obtain the fused linear acceleration, angular velocity, and underwater acoustic signal data from the data fusion module, construct sample data according to a predefined time window, and calculate the covariance matrix R using the sampled data i , and then perform eigenvalue decomposition on the covariance matrix R i to extract the main components of the signal and obtain the distribution of the signal subspace and the noise subspace;

[0053] (2) Fuse the second-order statistics between multiple channels and design a weighted fusion power spectrum accordingly:

[0054] ;

[0055] where represents the power response function of the j-th channel in the direction , and the weight is determined by the signal energy and its statistical characteristics of each channel. By weighting and fusing the information of multiple channels, the comprehensive beam pointing response is calculated;

[0056] (3) Traverse within a set angle range , calculate the above power spectrum for each angle, plot the results to form a beam pattern, and perform weighted fusion on the information in the aforementioned signal subspace and the weighted fusion power spectrum;

[0057] Specifically: Define an enhanced beam function:

[0058] ;

[0059] where is the steering vector in the desired direction , is the conjugate transpose of ;

[0060] (4) The actual obtained Regarded as the convolution result of the real target azimuth distribution with a point spread function PSF, that is:

[0061]

[0062] Using the deconvolution algorithm to perform iterative processing to solve the target distribution , and the iterative update formula is:

[0063] ;

[0064] Among them, is initialized by , is the flipped version of;

[0065] The final target direction estimation is the output result after the convergence of the iterative process:

[0066] ;

[0067] Among them, K is the number of steps when the iterative termination condition is satisfied. Through iterative update, the estimated target azimuth finally forms a clear peak in the correct direction, so as to accurately determine the target azimuth.

[0068] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0069] This solution effectively distinguishes underwater target signals through multi-sensor data fusion and intelligent signal processing. After being amplified and conditioned by the signal conditioning module, and creatively using algorithms such as Eigen-AMVDR to achieve accurate target direction estimation, combined with the RS485 bus and differential compression algorithm, it greatly reduces the data transmission volume and ensures real-time performance, meeting the requirements of long-distance seabed monitoring and real-time interaction; and combined with the design of the calibration module, with built-in anomaly detection, timeout retransmission and fault isolation mechanisms, it can give timely feedback and automatically compensate when data anomalies or equipment failures occur, ensuring the long-term stable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is the system block diagram described in the embodiment of the present invention;

[0071] Figure 2 is the circuit schematic diagram of setting the PGA amplification gain in the analog-to-digital conversion module in the embodiment of the present invention;

[0072] Figure 3 is the circuit schematic diagram of the signal conditioning module in the embodiment of the present invention;

[0073] Figure 4 This is a schematic flowchart of the method according to the embodiments of the present invention. Detailed implementation manners

[0074] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described below with reference to the accompanying drawings and embodiments. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention may be implemented in other ways different from those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0075] Currently, in underwater target detection and positioning, the following three key problems are mainly faced:

[0076] (1) The information dimension of a single sensor is limited, and it is difficult to meet the requirements of high-precision three-dimensional perception. Currently, most underwater target detection systems only rely on a single underwater acceleration sensor to collect signals in a certain direction, and cannot comprehensively describe the true motion state of the target in three-dimensional space. Especially in tasks such as target trajectory recognition and fine positioning, the single-axis data has a viewing blind area, which severely restricts the system performance;

[0077] (2) The underwater environment is dynamically complex, and the sensor errors are time-varying and difficult to compensate independently. The underwater environment is interfered by multiple factors such as temperature, pressure, flow velocity, and ocean clutter. Sensors are prone to problems such as zero drift, gain change, and transient anomalies during actual deployment. Traditional systems lack the ability of dynamic identification and joint calibration of these error sources; (3) The system robustness is poor and it cannot cope with local sensor failures or data anomalies. Existing solutions generally lack a redundancy mechanism. Once a single sensor fails or the data is strongly interfered, the overall perception function of the system will be paralyzed.

[0078] In view of the above problems, the present invention introduces multi-sensor data fusion technology, and fuses multi-source data in terms of time, space, and statistical features to achieve the following goals: improving three-dimensional perception ability: fusing underwater acoustic signals, linear acceleration, angular velocity, and temperature information to obtain a complete description of the target motion; enhancing the system robustness and anti-interference ability: through multi-channel comparison, dynamic weighting, and anomaly rejection, realizing "perceptual redundancy" and "fault-tolerant fusion" to ensure that the system can still operate stably even when some sensors fail.

[0079] Embodiment 1, an intelligent underwater acoustic signal processing system based on multi-sensor data fusion, such as Figure 1As shown, it includes a main control unit, a data processing module and a multi-sensor module connected to the main control unit. The multi-sensor module includes a hydrophone, an underwater acceleration sensor, an attitude sensor and a temperature sensor. The underwater acceleration sensor and the attitude sensor respectively adopt an ADXL335 three-axis accelerometer and an MPU6050 attitude sensor, which can monitor the movement in six degrees of freedom to maintain high sensitivity in small vibration detection. The data processing module communicates with the main control unit through an SPI digital interface. The main control unit includes a data fusion module, a bearing calculation module and a calibration module. The main control unit is connected to the upper computer through a communication module to achieve data transmission and remote monitoring, and display data such as the acceleration curve, bearing angle and environmental parameters of the underwater target at the upper computer end.

[0080] The data processing module includes a signal conditioning module and an analog-to-digital conversion module. Since the signals in the underwater environment are weak and vulnerable to noise interference, the analog signals need to be amplified and digitized by the signal conditioning module and the analog-to-digital conversion module, and then transmitted to the main control unit for data fusion processing. Specifically:

[0081] First, the hydrophone continuously outputs very weak analog signals due to its own state changes and transmits them to the data processing module for processing. The signal conditioning module amplifies and conditions the analog signals of the sensor, and the amplified and conditioned signals are sent to the analog-to-digital conversion module for digitization. At the same time, the main control unit continuously collects the data of the underwater acceleration sensor and the attitude sensor to determine the operating attitude of the device and ensure the safe operation of the device.

[0082] The signal conditioning module adopts a differential amplifier circuit design. Specifically, combined with Figure 3 as shown:

[0083] The differential amplifier circuit uses the differential amplifier OPA1632 to construct a variable-gain differential amplifier network, including a first OPA module and a second OPA module connected in series. The positive input terminal of the first OPA module is connected to resistor R1, and the negative input terminal is connected to resistor R2. Resistor R3 and capacitor C1 are connected in parallel between the positive input terminal and the negative output terminal of the first OPA module, and resistor R4 and capacitor C2 are connected in parallel between the negative input terminal and the positive output terminal. Similarly, the positive input terminal of the second OPA module is connected to resistor R5, the negative input terminal is connected to resistor R6, resistor R7 and capacitor C3 are connected in parallel between the positive input terminal and the negative output terminal of the second OPA module, and resistor R8 and capacitor C4 are connected in parallel between the negative input terminal and the positive output terminal of the second OPA module. Capacitor C5 is connected to the negative output terminal of the second OPA module, and capacitor C6 is connected to the positive output terminal. Moreover, resistor R9 and R10 are respectively connected to the negative output terminal and the positive output terminal of the second OPA module and grounded. In this embodiment, the OPA1632 is used to construct a fully differential amplifier. A symmetric bridge negative feedback structure is formed by the input resistors R1, R2 and the feedback resistors R3, R4. Compared with the feedback design of traditional inverting amplifiers, the number of matching resistors is reduced by about 50%, simplifying the design and improving the system consistency. Its differential gain is approximately G = 2R3 / R1, with higher gain adjustment sensitivity, which is beneficial to achieving higher-precision gain control under the condition of the same resistor error. In addition, capacitors C1, C2, C3, and C4 are used for filtering and decoupling to eliminate high-frequency noise and stabilize the power supply voltage.

[0084] Traditional differential amplifier circuits usually use operational amplifiers + feedback resistors to achieve. The differential amplifier OPA1632 used in the present invention has optimized matching and bias circuits inside, providing better CMRR (common-mode rejection ratio) and distortion performance. In classic differential amplifier circuits, the signal gain is usually directly set through resistors. In the circuit design of this solution, coupling capacitors (C1, C2, C3, C4) are used at the input end, which can filter out DC offsets and avoid drift problems caused by DC amplification. In addition, the input end (R1 - R4) of this circuit has a high impedance, which can effectively reduce the load impact on the previous-stage circuit. There are R9 and R10 as termination matching resistors at the output end of this circuit, while classic differential amplifier circuits usually do not require additional termination resistors. This design helps to optimize the impedance matching of differential signals, reduce signal reflection, and improve signal integrity.

[0085] The analog-to-digital conversion module uses a multi-channel synchronous sampling 24-bit Δ-Σ analog-to-digital converter circuit (ADS1292). The analog-to-digital conversion module incorporates a programmable gain amplifier (PGA), an internal reference, and an on-board oscillator. The analog-to-digital conversion module has two input signals: MuxP (positive input) and MuxN (negative input), combined with Figure 2As shown, the signal OUT+ terminal is connected to the positive input terminal of signal No. 1 of ADS1292, and the signal OUT- terminal is connected to the negative input terminal of signal No. 1 of ADS1292. The + / - input terminals of signal No. 2 of ADS1292 are short - circuited and connected to the +2.5V power supply terminal. Resistors Ra and Rs are used to set the gain of the amplifier. The PGA output is filtered by an RC filter before entering the ADC. The filter consists of an internal resistor RS = 2kΩ and an external capacitor CFILTER (typical value is 4.7nF). The larger the capacitance value, the worse the total harmonic distortion (THD) performance. The internal RS resistor is accurate to 15%, so the actual bandwidth will vary.

[0086] After the digital signal enters the main control unit, data pre - processing is first performed, and then data fusion processing is carried out. In this embodiment, a differential compression algorithm is adopted to reduce the amount of transmitted data by 60%. To achieve multi - sensor data fusion, it is necessary to allocate independent timer resources in the main control unit to accurately timestamp each acquisition of linear acceleration, angular velocity, and underwater acoustic signal data to ensure that the data of each sensor is aligned on the same time axis. Through the DMA (Direct Memory Access) mechanism inside the main control unit, zero - delay transmission is achieved to optimize the RS485 bus. The data stream is stored in the SRAM in the form of a buffer with a fixed length, which can reduce the CPU burden and provide batch data for subsequent algorithm processing. The data pre - processing of the main control unit includes:

[0087] 1) Spatiotemporal alignment: By uniformly timestamping the sampled data of different sensors, the time synchronization of multi - source data is achieved;

[0088] 2) Static reference calibration: Collect data in a stationary state during the system startup phase to construct a linear calibration model;

[0089] 3) Temperature compensation: Based on the real - time collected temperature data and combined with the temperature drift models of each sensor, the measured values of underwater acoustic signals, angular velocity, and linear acceleration are corrected to reduce the zero - point offset and gain error caused by temperature changes.

[0090] The present invention combines underwater acceleration, attitude, temperature sensors and hydrophones for multi-source information fusion, compensates the linear acceleration data for attitude, eliminates the additional linear acceleration components caused by attitude changes, and completes the fused underwater acoustic signal. The linear acceleration and angular velocity are sent to the azimuth calculation module, which combines the Eigen-AMVDR algorithm to accurately estimate the direction of the target. The motion state data and azimuth angle of the target are calculated through the azimuth calculation module, so that the signal in the desired direction is distortion-free and the output power remains unchanged, and the noise in the non-output desired direction is minimized, making the final beam output power minimum, thereby obtaining the target azimuth. Among them, the fusion process is divided into two parts: inertial data fusion and underwater acoustic signal fusion. The former is used for attitude / linear acceleration compensation, and the latter is used for underwater acoustic array direction correction. Specifically:

[0091] Inertial data fusion:

[0092] 1) Attitude calculation: Fuse the linear acceleration and angular velocity data, and use the complementary filtering method to calculate the current attitude angles (Pitch, Roll, Yaw) of the device;

[0093] 2) Attitude compensation: Based on the current attitude angles, construct a rotation matrix to compensate the three-axis linear acceleration data and eliminate the pseudo linear acceleration components caused by the attitude changes of the device;

[0094] 3) Weighted fusion: Dynamically allocate weights according to the historical stability and current deviation of the sensors. The higher the stability and the smaller the deviation, the higher the weight will be assigned;

[0095] Calculate the weighted average of multiple channels as the fusion result; the so-called weighted average is the average method of weighted processing according to the weights of each value;

[0096] 4) Anomaly rejection: When the data of a certain channel significantly deviates from the statistical range of the other sensors, dynamically reduce its weight or temporarily reject it to enhance the robustness of the system;

[0097] 5) Three-dimensional vector synthesis: Synthesize the corrected X / Y / Z three-axis linear acceleration data through the Euclidean norm to characterize the overall perturbation amplitude of the device in three-dimensional space and reflect its current motion trend or stability state.

[0098] Underwater acoustic signal fusion:

[0099] After the underwater acoustic signal is conditioned and analog-to-digital converted by the preprocessing module, the system uses the rotation matrix constructed in the attitude compensation step to perform real-time correction on its receiving direction, compensates for the pointing deviation of the hydrophone array caused by attitude perturbation, and ensures that the array beam is consistent with the reference direction. Finally, the corrected underwater acoustic signal, linear acceleration, and angular velocity data are synchronously encapsulated under the unified time stamp and attitude reference and enter the azimuth calculation module.

[0100] It should be particularly noted that: Since this embodiment uses a fixed underwater device, all the data collected by the linear acceleration and attitude sensors are from the state of the device body, rather than the acceleration of external targets. The fused angular velocity and the triaxial acceleration data after attitude and temperature compensation are used to assist in beam direction correction, coordinate mapping in target direction estimation, as well as the detection and compensation processing of the attitude change of the device itself, ensuring the stability of the reference system for underwater acoustic signal processing and the accuracy of direction estimation.

[0101] The data fusion result is the corrected triaxial acceleration + angular velocity + underwater acoustic signal spatio-temporal mapping data + temperature data obtained after temperature compensation, attitude compensation, weighted fusion, and outlier rejection, which is used to characterize the motion state of the underwater device, target azimuth perception, and environmental response in a unified reference coordinate system. The format of the fused data frame is shown in Table 1. Each frame is in a fixed-length format and is transmitted in hexadecimal form.

[0102] Table 1 Format of the fused data frame

[0103] Byte position Number of bytes Field name Description 0-1 2B Data frame header Fixed identifier (0xAA55) for frame start recognition 2-5 4B Timestamp Microsecond timer, accurately recording the data acquisition time 6-11 6B Underwater acoustic signal Underwater acoustic data 12-17 6B Three-axis acceleration X / Y / Z three-axis acceleration 18-23 6B Three-axis angular velocity Pitch, Roll, Yaw angular velocity 24-25 2B Temperature data Temperature value 26 1B Status flag bit0: Calibration status, bit1: Sensor anomaly 27-28 2B Checksum CRC-16 or checksum for data integrity verification 29-30 2B Data frame tail Fixed frame tail identifier (0X55AA)

[0104] The data fusion result is input into the azimuth calculation module. Among them, the corrected underwater acoustic signal is the main input of the azimuth calculation module, which is used to construct the direction covariance matrix and is the core basis for calculating the target azimuth angle. Combining with the Eigen-AMVDR algorithm for target direction estimation, the target azimuth angle and motion state parameters are finally output. The motion state parameters include the device attitude angle and the synthetic acceleration, which are used to assist in target dynamic behavior analysis and system attitude compensation. The principle of the azimuth calculation module is as follows:

[0105] First, calculate the covariance matrix of the received signal, which contains the spatial characteristics of the array received signal. Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors. The eigenvectors corresponding to the larger eigenvalues correspond to the signal subspace, while the eigenvectors corresponding to the smaller eigenvalues correspond to the noise subspace. Use the eigenvectors to project the received signal onto the signal subspace and the noise subspace, thereby separating the signal and noise components.

[0106] Apply the Eigen-AMVDR algorithm in the signal subspace to calculate the optimal weight vector W, so that the gain in the target direction is maximized while suppressing interference and noise in other directions. This algorithm can more accurately separate the signal and noise and improve the resolution of the beamformer. Specifically:

[0107] (1) Obtain the fused linear acceleration, angular velocity, and underwater acoustic signal data from the data fusion module, construct sample data according to the predefined time window, and calculate the covariance matrix R using the sampled data i , that is

[0108] ;

[0109] in, Respectively The output vectors of the underwater accelerometer, attitude sensor, and hydrophone are captured in a snapshot, where i = 1, 2, 3, ..., representing different sensor types. This step collects data from the multi-channel underwater accelerometer, attitude sensor, and hydrophone, extracting the signal covariance information from the sensor array. This information reflects the spatial distribution characteristics of the signals and provides basic data for target direction estimation.

[0110] Covariance matrix R i Perform eigenvalue decomposition and get

[0111] ;

[0112] in, is the eigenvector matrix, is the corresponding diagonal matrix of eigenvalues. Through eigendecomposition, the main components of the signal are extracted, and the distribution of the signal subspace and the noise subspace is obtained. The decomposed eigenvalues and eigenvectors can effectively distinguish between signals and noise, improving target detection capabilities.

[0113] (2) Based on the traditional method of using only the information of a single route acceleration sensor channel for direction estimation, this embodiment further fuses the second-order statistics between multiple channels (including linear acceleration, angular velocity and underwater acoustic signal) and designs a weighted fusion power spectrum based on this:

[0114] ;

[0115] in, Indicates that the jth channel is in the direction The power response function on The energy and statistical characteristics of the signals in each channel are determined by weighted fusion of information from multiple channels, calculating the comprehensive beam pointing response and combining data from multiple sensors to optimize direction estimation, reduce the impact of individual channel noise, and improve robustness.

[0116] (3) Traverse within the set angle range (such as -90° to +90°) , calculate the above power spectrum for each angle , and the result can be plotted to form a beam diagram. The information in the aforementioned signal subspace is weightedly fused with the weighted fusion power spectrum. Specifically, an enhanced beam function is defined:

[0117] ;

[0118] in, Is the expected direction The steering vector on is the conjugate transpose of. This step uses the signal subspace for projection and combines the enhanced beam function to improve the contrast of direction estimation. By enhancing the signal energy, the resolution ability of the target azimuth is improved, making the target angle easier to be recognized. It has both the advantage of the signal subspace suppressing noise and introduces the rich spatial information provided by the second-order statistics, thus making the main lobe narrower and the side lobes lower.

[0119] (4)The actually obtained can be regarded as the convolution result of the true target azimuth distribution and a point spread function (PSF), that is:

[0120] ;

[0121] This formula indicates that the enhanced beam function is equivalent to the target azimuth distribution, indicating that the original direction estimation may be extended or blurred, and further optimization is needed to obtain clearer target azimuth information. Using the deconvolution algorithm to perform iterative processing to solve for a sharper target distribution . The iterative update formula is:

[0122] ;

[0123] where is initialized by , and is the flipped version of.

[0124] The final target direction estimation is the output result after the iteration process converges:

[0125] ;

[0126] where K is the number of steps when the iteration termination condition is satisfied. Through iterative update, finally, the estimated target azimuth forms a clear peak in the correct direction, thus accurately determining the target azimuth.

[0127] After several iterations, will show obvious sharp peaks, and these peaks are the direction estimation results of the target.

[0128] It can be seen that the Eigen-AMVDR algorithm removes the noise subspace and only leaves the signal subspace in the output covariance matrix. While reducing the output power of the noise, the output energy in the desired direction remains basically unchanged, thereby improving the output signal-to-noise ratio and obtaining a sharper direction estimation spectrum.

[0129] In addition, considering that in the actual underwater environment, abnormal situations such as sensor damage or sudden noise spikes may occur, the data processing module introduces a sliding window statistical anomaly detection mechanism to give an alarm in time or automatically restart the sensor calibration process when data anomalies are found, so as to ensure the overall stability of the system. The calibration module dynamically compensates for the errors of the sensors to ensure the data accuracy and stability of the system during long-term underwater operations.

[0130] Real-time dynamic compensation steps: During the operation of the system, the output data of the sensors are continuously collected and compared in real time. Using the sliding window technology, the linear acceleration data collected each time are analyzed, and the mean and variance are calculated. If the data of multiple consecutive windows exceed the set tolerance range, it is regarded as a deviation. When a deviation is detected, the system automatically triggers "secondary calibration": a low-pass filter is applied to filter out the sudden noise, the new bias and gain values are recalculated, and these parameters are dynamically loaded into the calibration module of the system for calibration. The calibrated data will continue to be used in the real-time processing chain to ensure the accuracy of the signals output by the system.

[0131] The communication module of this embodiment uses the RS485 bus standard for data transmission. The core controller of the main control unit uses a single-chip microcomputer (STM32F103), which is connected to the USART interface of the main control unit through the TX, RX, and DE / RE pins to complete the signal level conversion of the physical layer. In order to ensure the signal quality, a 120Ω terminal matching resistor needs to be added at both ends of the bus. In addition, in order to improve the stability of the bus idle state, a bias resistor is added to the bus A+ / B- to ensure that the bus is at a known level when it is idle. At the RS485 level, differential signal transmission is used, and the anti-interference ability is relatively strong. Since real-time data transmission is required in the present invention, a relatively high baud rate (115200bps) is used to minimize the bit error rate while meeting the real-time requirements.

[0132] During specific operation, after the system starts up, the underwater acceleration sensor, temperature and attitude sensors, etc. are first placed in a static and stable state to collect a segment of reference data. By taking the average value of multiple sampling points, the initial zero offset and gain error of each sensor are calculated. The initial calibration results are stored in the non-volatile memory as the initial reference for subsequent dynamic compensation. During the operation of the system, the output changes of the sensors are continuously detected using the real-time data stream. When a deviation from the long-term statically collected data is detected, the system automatically enters the secondary calibration mode, performs low-pass filtering on the latest sampling data to remove instantaneous noise, and then calculates new calibration coefficients. Through the integrated temperature sensor, the ambient temperature data is collected in real time, and the relationship curve between the sensor output and temperature is established to dynamically adjust the calibration parameters. The data of each sensor is compared and fused through a fusion algorithm. By comparing the consistency of the outputs of each sensor, abnormal data is identified, and the weights of each sensor are adjusted according to the statistical model to minimize the error. The attitude information of the device is obtained in real time using the attitude sensor, and the linear acceleration data is corrected for attitude. A periodic calibration task is set to run with a timed interrupt in the main control unit to ensure that the calibration parameters are always in the optimal state during long-term operation. When the main control unit detects abnormal sampling data, the system will automatically trigger the alarm mechanism, feedback the abnormal information to the host computer, and reset it to the initial calibration value. Through the above specific implementation process of the calibration module and error compensation, this solution can correct the zero drift, gain error of the sensor, and the deviation caused by factors such as temperature and pressure in real time in the complex and changeable underwater environment, ensuring the high precision and stability of the output data.

[0133] In this embodiment, each module is integrated in an underwater device, effectively ensuring the stable operation of the system in the complex dynamic underwater environment. The entire system has a high degree of integration and reliability and can still work normally in harsh environments. In practical applications, the system can be deployed in fields such as seabed resource exploration, ocean engineering monitoring, and ocean security prevention. By real-time monitoring the vibration signals and motion states of underwater targets, accurate azimuth information is provided to help the operator judge the target position and motion trajectory, providing reliable data support for subsequent action decisions.

[0134] Embodiment 2. The signal processing method of the intelligent underwater acoustic signal processing system based on multi-sensor data fusion proposed in Embodiment 1, combined with Figure 4 as shown, includes the following steps:

[0135] Step A. Data acquisition and preprocessing:

[0136] The system collects the three-axis acceleration and angular velocity information of the device body in the underwater environment through underwater acceleration sensors (such as ADXL335) and attitude sensors (such as MPU6050) to perceive the platform perturbation trend and attitude stability. Meanwhile, the underwater acoustic signals are collected in real time by an array of hydrophones for subsequent direction estimation and sound source identification. The analog signals output by all sensors are amplified by an amplifier circuit and then converted into digital signals, followed by signal preprocessing, and then uniformly input into the main control unit, laying a foundation for subsequent fusion processing and beamforming.

[0137] Step B: Data fusion and filtering:

[0138] Fuse the data from multiple sensors (hydrophones, underwater acceleration sensors, attitude sensors, temperature sensors, etc.) through a data fusion module, and remove noise through filtering to ensure the accuracy of the data. Use spatio-temporal alignment coding to synchronize the data so that the multi-sensor data can work accurately in coordination.

[0139] Step C: Position and direction calculation:

[0140] Combined with the azimuth calculation module, use the Eigen-AMVDR algorithm. By calculating the covariance matrix of the signal, extract the eigenvectors of the signal, and then estimate the direction and azimuth angle of the target. And transmit the processed data through the RS485 bus.

[0141] Step D: Error calibration and compensation:

[0142] Combined with the calibration module, perform reference calibration through static data acquisition to obtain the zero offset and gain error of the sensor. During the actual operation process, the system monitors the data changes in real time and performs calibration through a dynamic compensation mechanism to correct the errors caused by factors such as temperature and pressure.

[0143] Step E: Data monitoring and display: Finally, the processed data is displayed and monitored through the upper computer. The user can view the motion state, acceleration curve, and azimuth information of the target in real time. The upper computer can communicate with the device bidirectionally through the serial port for operations such as configuration adjustment and parameter modification.

[0144] Specifically, in step C, when performing position and azimuth calculation, the following method is specifically adopted:

[0145] (1) Obtain the fused linear acceleration, angular velocity, and underwater acoustic signal data from the data fusion module, construct sample data according to a predefined time window, and calculate the covariance matrix R using the sampled data i , that is

[0146] ;

[0147] Among them, n This step collects data from multi-channel underwater acceleration sensors, attitude sensors, and hydrophones, extracts the signal covariance information of the sensor array, reflects the spatial distribution characteristics of the signals, and provides basic data for the direction estimation of the target.

[0148] Perform eigenvalue decomposition on the covariance matrix R to obtain

[0149] ;

[0150] where is the eigenvector matrix, is the corresponding eigenvalue diagonal matrix. Through eigen-decomposition, the main components of the signals are extracted, and the distributions of the signal subspace and the noise subspace are obtained. Through the decomposed eigenvalues and eigenvectors, signals and noises can be effectively distinguished, and the target detection ability can be improved.

[0151] (2) In addition to using the information of a single channel, in this embodiment, the second-order statistics between channels are further calculated, and a weighted fusion power spectrum is designed based on this;

[0152] ;

[0153] where represents the directional response contributed by the j-th channel, and the weight is determined by the signal energy and its statistical characteristics of each channel. By weighted-fusing the information of multiple channels, the comprehensive beam pointing response is calculated, combining the data of multiple sensors, optimizing the direction estimation, reducing the influence of the noise of a single channel, and improving the robustness.

[0154] (2) Weightedly fuse the information in the aforementioned signal subspace with the weighted fusion power spectrum. Specifically: Define an enhanced beam function:

[0155] ;

[0156] where is the steering vector in the desired direction . This step uses the signal subspace for projection, and combines the original statistics to improve the contrast of the direction estimation. By enhancing the signal energy, the resolution ability of the target azimuth is improved, making the target angle easier to be recognized. It has both the advantage of the signal subspace suppressing noise and introduces the rich spatial information provided by the second-order statistics, thus making the main lobe narrower and the side lobes lower.

[0157] (3) The actually obtained can be regarded as the true target azimuth distribution The convolution result with a point spread function (PSF), i.e.:

[0158] .

[0159] This formula indicates that the enhanced beam function is equivalent to the target azimuth distribution, suggesting that the original direction estimation may have expansion or ambiguity and needs further optimization to obtain clearer target azimuth information. Using the deconvolution algorithm to perform iterative processing to solve for a sharper target distribution .

[0160] The iterative update formula is:

[0161] ;

[0162] where is initialized by , and is the flipped version of .

[0163] The final target direction estimation is the output result after the iteration process converges:

[0164] ;

[0165] where K is the number of steps when the iteration termination condition is met. Through iterative update, finally, the estimated target azimuth forms a clear peak in the correct direction, thus accurately determining the target azimuth. After several iterations, will show obvious sharp peaks, and these peaks are the direction estimation results of the target.

[0166] It can be seen from the formula that the Eigen-AMVDR algorithm removes the noise subspace and only leaves the signal subspace in the output covariance matrix. While reducing the noise output power, the output energy in the desired direction remains basically unchanged, thereby improving the output signal-to-noise ratio and obtaining a sharper direction estimation spectrum.

[0167] This solution provides a brand-new solution for underwater target detection by integrating high-precision sensors, multi-level data processing, intelligent calibration, and efficient communication technologies, effectively improving the detection accuracy and system stability, and having broad application potential.

[0168] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent underwater acoustic signal processing system based on multi-sensor data fusion, characterized in that, It includes a main control unit, a data processing module and a multi-sensor module connected to the main control unit. The multi-sensor module includes a hydrophone, an underwater acceleration sensor, an attitude sensor and a temperature sensor, which respectively collect underwater acoustic signals, linear acceleration signals, angular velocity signals and temperature signals; The data processing module includes a signal conditioning module and an analog-to-digital conversion module. The analog signals collected by the sensor module are amplified and conditioned by the signal conditioning module, and the amplified and conditioned signals are sent to the analog-to-digital conversion module for digitization; The main control unit includes a data fusion module, a bearing calculation module and a calibration module. The data fusion module receives the output information of the data processing module for data fusion, and inputs the data fusion result into the bearing calculation module. Combining with the Eigen-AMVDR algorithm, beamforming and target direction estimation are carried out, and finally the target bearing angle and motion state parameters are output, and the output accuracy is ensured in combination with the calibration module; When the data fusion module performs fusion processing, it includes inertial data fusion and underwater acoustic signal fusion. Specifically: Inertial data fusion: First, fuse the linear acceleration and angular velocity signals, and use the complementary filtering method to solve the current attitude angle; construct a rotation matrix based on the current attitude angle to perform attitude compensation on the linear acceleration signal; perform weighted fusion in combination with dynamic weight allocation, and calculate the multi-channel weighted average value as the fusion result; finally, synthesize the corrected linear acceleration data; Underwater acoustic signal fusion: Use the rotation matrix to perform real-time correction on its receiving direction. The corrected underwater acoustic signal is synchronously encapsulated with the data after inertial data fusion under the same timestamp and attitude reference, and enters the bearing calculation module; The bearing calculation module constructs a covariance matrix for the received signal, performs eigenvalue decomposition on the covariance matrix, and obtains eigenvalues and eigenvectors; Eigenvalue The large eigenvectors correspond to the signal subspace, while the small eigenvectors correspond to the noise subspace. Use the eigenvectors to project the received signal onto the signal subspace and the noise subspace to separate the signal and noise components; Apply the Eigen-AMVDR algorithm in the signal subspace to calculate the optimal weight vector W, so that the gain in the target direction is maximized, and at the same time, the interference and noise in other directions are suppressed. Specifically: (1) Obtain the fused linear acceleration, angular velocity, and underwater acoustic signal data from the data fusion module, construct sample data according to a predefined time window, and calculate the covariance matrix R using the sampled data i , and then for the covariance matrix R i perform eigenvalue decomposition, extract the main components of the signal, and obtain the distribution of the signal subspace and the noise subspace; (2)Fuse the second-order statistics between multiple channels and design a weighted fusion power spectrum accordingly: ； Among them, represents the power response function of the j-th channel in the direction The weights are determined by the signal energy and its statistical characteristics of each channel. By weighted fusion of the information of multiple channels, the comprehensive beam pointing response is calculated; (3) Traverse within the set angular range and calculate the above-mentioned power spectrum for each angle Plotting the results can form a beam pattern, and the information in the aforementioned signal subspace is weighted and fused with the weighted fusion power spectrum; Specifically: Define an enhanced beam function: ； Among them, is the steering vector in the desired direction and is the conjugate transpose of . (4) Actually obtained regarded as the true target azimuth distribution the convolution result with a point spread function PSF, that is: ； Use the deconvolution algorithm to perform iterative processing to solve the target distribution , and the iterative update formula is: ； Among them, Initialized by Initialized, is the flipped version of Final target direction estimation That is, the output result after the iteration process converges: ； where K is the number of steps when the iteration termination condition is met. Through iterative update, finally a clear peak is formed in the correct direction, so as to accurately determine the target orientation; ​ The main control unit is connected to the upper computer through a communication module to realize data transmission and remote monitoring.

2. The intelligent underwater acoustic signal processing system based on multi-sensor data fusion according to claim 1, characterized in that: The calibration module is used to achieve real-time dynamic compensation: During the operation of the system, continuously collect the output data of each sensor and perform real-time comparison. Use the sliding window technology to analyze the linear acceleration data collected each time, and calculate its mean and variance; if the data of multiple consecutive windows exceeds the set tolerance range, it is regarded as a deviation; when a deviation is detected, based on the calibration module, apply a low-pass filter to filter out the sudden noise, recalculate the new bias and gain values, and dynamically load the new bias and gain value parameters into the calibration module for calibration.

3. The intelligent underwater acoustic signal processing system based on multi-sensor data fusion according to claim 1, characterized in that: The signal conditioning module is designed with a differential amplifier circuit. The differential amplifier circuit uses the differential amplifier OPA1632 to construct a variable-gain differential amplification network, which includes a series-connected first OPA module and second OPA module. The positive input terminal of the first OPA module is connected to resistor R1, the negative input terminal is connected to resistor R2. Resistors R3 and capacitor C1 are connected in parallel between the positive input terminal and the negative output terminal of the first OPA module, and resistors R4 and capacitor C2 are connected in parallel between the negative input terminal and the positive output terminal. Similarly, the positive input terminal of the second OPA module is connected to resistor R5, the negative input terminal is connected to resistor R6. Resistors R7 and capacitor C3 are connected in parallel between the positive input terminal and the negative output terminal of the second OPA module, and resistors R8 and capacitor C4 are connected in parallel between the negative input terminal and the positive output terminal of the second OPA module. Capacitor C5 is connected to the negative output terminal of the second OPA module, and capacitor C6 is connected to the positive output terminal. And resistors R9 and R10 are respectively connected to the negative output terminal and the positive output terminal of the second OPA module and grounded.

4. The intelligent underwater acoustic signal processing system based on multi-sensor data fusion according to claim 1, characterized in that: After the data processed by the data processing module is transmitted to the main control unit, data preprocessing is first performed. The preprocessed data is then fused by the data fusion module. When performing data preprocessing, it specifically includes: 1) Space-time alignment: Mark the sampling data of different sensors through a unified time stamp to achieve time synchronization of multi-source data; 2) Static reference calibration: Collect data in a static state during the system startup phase to construct a linear calibration model; 3) Temperature compensation: Based on the real-time collected temperature data and combined with the temperature drift models of each sensor, correct the measured values of underwater acoustic signals, angular velocity, and linear acceleration.

5. The method of the intelligent underwater acoustic signal processing system based on multi-sensor data fusion according to claim 1, characterized in that: It includes the following steps: Step A, Data acquisition and preprocessing: The multi-sensor module collects underwater-related signals. The analog signals collected by the sensor module are amplified and conditioned by the signal conditioning module, and the amplified and conditioned signals are sent to the analog-to-digital conversion module for digitization, and then uniformly input to the main control unit for signal preprocessing; Step B, Data fusion and filtering: The data preprocessed by the main control unit undergoes multi-source information fusion through the data fusion module, performs attitude compensation on the linear acceleration signal, and removes noise through filtering; Step C, Position and direction calculation: In combination with the azimuth calculation module, use the Eigen-AMVDR algorithm, calculate the covariance matrix of the signal, extract the eigenvectors of the signal, and then estimate the direction and azimuth angle of the target; Step D, Error calibration and compensation: In combination with the calibration module, perform reference calibration through static data acquisition to obtain the zero offset and gain error of the sensor. During the actual operation process, calibrate by real-time monitoring of data changes and combining with the dynamic compensation mechanism; Step E, Data monitoring and display: The finally processed data is displayed and monitored through the upper computer.

6. The method of the intelligent underwater acoustic signal processing system based on multi-sensor data fusion according to claim 5, characterized in that: In step A, when the main control unit performs signal preprocessing, it includes the following steps: 1) Space-time alignment: Mark the sampling data of different sensors through a unified time stamp to achieve time synchronization of multi-source data; 2) Static reference calibration: Collect data in a static state during the system startup phase to construct a linear calibration model; 3) Temperature compensation: Based on the real-time collected temperature data and combined with the temperature drift models of each sensor, the measured values of the underwater acoustic signal, angular velocity, and linear acceleration are corrected to reduce the zero-point offset and gain error caused by temperature changes.

7. The method of the intelligent underwater acoustic signal processing system based on multi-sensor data fusion according to claim 5, characterized in that: In step C, when performing azimuth calculation, the principle is as follows: The azimuth calculation module constructs a covariance matrix for the received signal, performs eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors; the eigenvector corresponding to the larger eigenvalue corresponds to the signal subspace, while the eigenvector corresponding to the smaller eigenvalue corresponds to the noise subspace. The received signal is projected onto the signal subspace and the noise subspace using the eigenvectors to separate the signal and noise components; In the signal subspace, the Eigen-AMVDR algorithm is applied to calculate the optimal weight vector W, so that the gain in the target direction is maximized while suppressing interference and noise in other directions. Specifically: (1) Obtain the fused linear acceleration, angular velocity, and underwater acoustic signal data from the data fusion module, construct sample data according to a predefined time window, and calculate the covariance matrix R using the sampled data i , and then for the covariance matrix R i perform eigenvalue decomposition, extract the main components of the signal, and obtain the distribution of the signal subspace and the noise subspace; (2) Fuse the second-order statistics between multiple channels and design a weighted fusion power spectrum accordingly: ； Among them, represents the power response function of the j-th channel in the direction The weight is determined by the signal energy and its statistical characteristics of each channel. By weighted fusion of the information of multiple channels, the comprehensive beam pointing response is calculated; (3) Traverse within the set angular range and calculate the above-mentioned power spectrum for each angle Then, plotting the results can form a beam pattern. The information in the aforementioned signal subspace is weighted and fused with the weighted fusion power spectrum; Specifically: Define an enhanced beam function: ； Among them, is the steering vector in the desired direction , and is the conjugate transpose of (4) Actually obtained regarded as the true target azimuth distribution the convolution result with a point spread function PSF, that is: ； Use the deconvolution algorithm to perform iterative processing to solve the target distribution , and the iterative update formula is: ； Among them, Initialized by Initialized For The flipped version of; Final target direction estimation That is, the output result after the convergence of the iterative process: ； where K is the number of steps when the iteration termination condition is satisfied. Through iterative update, finally, a clear peak is formed in the correct direction, so as to accurately determine the target azimuth. in the correct direction, thus accurately determining the target azimuth.

8. The method of the intelligent underwater acoustic signal processing system based on multi-sensor data fusion according to claim 5, characterized in that: In step D, it is specifically implemented based on the following principle: During the operation of the system, the output data of each sensor is continuously collected and compared in real time. The sliding window technique is used to analyze the linear acceleration data collected each time, and its mean and variance are calculated; if the data of multiple consecutive windows exceeds the set tolerance range, it is regarded as a deviation; when a deviation is detected, based on the calibration module, a low-pass filter is applied to filter out the sudden noise, the new bias and gain values are recalculated, and the new bias and gain value parameters are dynamically loaded into the calibration module for calibration.

Images