Underwater target intelligent electromagnetic detection method and system

Abstract

The invention relates to an underwater target intelligent electromagnetic detection method and system, and belongs to the technical field of geophysical detection and object detection. The electromagnetic detection subsystem collects underwater electromagnetic signals; the electromagnetic data preprocessing unit decomposes an underwater electromagnetic data sample into a time domain signal sequence, a frequency domain power spectrum signal sequence and a time frequency power spectrogram; the time domain signal deep learning network identifies whether an underwater target exists according to the time domain signal sequence; the frequency domain power spectrum signal deep learning network identifies whether an underwater target exists according to the frequency domain power spectrum signal sequence; the time-frequency power spectrum signal deep learning network identifies whether an underwater target exists or not according to the time-frequency power spectrum; and according to the recognition result of the time-domain signal deep learning network on the underwater target, the recognition result of the frequency-domain power spectrum signal deep learning network on the underwater target and the recognition result of the time-frequency power spectrum signal deep learning network on the underwater target, performing comprehensive judgment to generate a final underwater target recognition result marked by a time-frequency power spectrogram.

Description

Technical Field

[0001] This application relates to the field of geophysical exploration and object detection technology, and in particular to an intelligent electromagnetic detection method and system for underwater targets. Background Technology

[0002] Acoustic detection has long been the primary method for detecting underwater targets. However, with continuous improvements in noise reduction technology, the acoustic radiation signals of underwater targets have significantly weakened, gradually approaching the level of ocean background noise. Acoustic detection alone can no longer meet the needs of underwater target detection, necessitating the development of non-acoustic detection technologies. The main non-acoustic detection technologies include optical imaging detection, lidar detection, infrared detection, and electromagnetic detection. Optical imaging detection is suitable for clear waters, but its performance is limited by lighting conditions and water transparency, restricting its long-range applications. Lidar detection suffers from significant signal attenuation and scattering in turbid seawater, limiting its long-range applications. Infrared detection utilizes temperature differences for detection, a passive method with some concealment, but its accuracy is easily affected by seawater stratification. Electromagnetic detection identifies target features and locations by detecting changes in the electromagnetic field caused by underwater targets, offering advantages such as low visibility, strong anti-interference capabilities, high hydrological adaptability, and high positioning accuracy. Compared to other non-acoustic detection technologies, electromagnetic detection demonstrates stronger applicability and technological advantages in complex marine environments, and can be widely applied to underwater target detection, identification, and location.

[0003] Currently, underwater target electromagnetic detection technology has made significant progress, mainly developing into fixed electromagnetic detection systems based on wired networks, mobile electromagnetic detection systems based on underwater mobile platforms, and surface-mounted towed electromagnetic detection systems. Fixed electromagnetic detection systems based on wired networks are primarily developed for near-shore underwater target detection needs. These systems consist of multiple nodes fixed on the seabed to collect electric, magnetic, and other physical field data. These nodes are connected to a shore-based monitoring center via a wired network, enabling real-time detection of the electromagnetic characteristics of underwater targets. While these systems are built on high-speed wired networks and can achieve real-time transmission of electromagnetic detection data, they suffer from high deployment costs, high difficulty, poor mobility, and limited detection range. Mobile electromagnetic detection systems based on underwater mobile platforms are primarily developed to meet the needs for mobile and wide-area detection of underwater targets. These systems utilize AUVs, Underwater gliders and other underwater mobile platforms serve as carriers, carrying electric and magnetic field detection systems to achieve mobile and wide-range detection of the electromagnetic characteristics of underwater targets. The advantages of this system are flexible deployment, good mobility, and a large detection range. However, it is limited by low-bandwidth wireless communication methods such as underwater acoustic communication, making it difficult to achieve real-time transmission of underwater target electromagnetic detection data. Surface-mounted towed electromagnetic detection systems are built on high-speed wired networks and achieve large-scale real-time mobile detection of underwater targets through the towing of surface vessels. This combines the advantages of fixed electromagnetic detection systems based on wired networks and mobile electromagnetic detection systems based on underwater mobile platforms.

[0004] From the perspective of underwater target electromagnetic signal identification methods, the marine electromagnetic environment is complex, containing various natural electromagnetic field signals generated by ocean waves, currents, tides, and the seabed medium induced in the Earth's magnetic field. Simultaneously, there are various artificial electromagnetic field signals generated by marine engineering activities, submarine cables, and nearshore power systems. These electromagnetic field signals couple with the weak electromagnetic field signals of underwater targets, making underwater target electromagnetic signal identification difficult. Deep learning, transfer learning, and other artificial intelligence technologies have made significant progress in image recognition, speech recognition, and natural language processing, providing technical guidance for the intelligent detection and identification of underwater target electromagnetic signals. Currently, there is no intelligent identification method for underwater target electromagnetic signals based on deep learning or other artificial intelligence algorithms. Summary of the Invention

[0005] This invention addresses at least one of the technical problems in related technologies by proposing an intelligent electromagnetic detection method and system for underwater targets based on artificial intelligence algorithms, thereby solving the problem of poor accuracy in detecting electromagnetic signals of underwater targets in existing technologies.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a system comprising: A method for intelligent electromagnetic detection of underwater targets includes the following steps: S1: Collect underwater electromagnetic signals to obtain underwater electromagnetic data and electromagnetic data samples; S2: Decompose the electromagnetic data sample into a time-domain signal sequence, a frequency-domain power spectrum signal sequence, and a time-frequency power spectrum diagram; S3: Input the time-domain signal sequence into a time-domain signal deep learning network to identify whether there is an underwater target in the time-domain signal sequence; input the frequency-domain power spectrum signal sequence into a frequency-domain power spectrum signal deep learning network to identify whether there is an underwater target in the frequency-domain power spectrum signal sequence; input the time-frequency power spectrum graph into a time-frequency power spectrum signal deep learning network to identify whether there is an underwater target in the time-frequency power spectrum graph. S4: Based on the underwater target recognition results of the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network, a comprehensive decision is made to generate the final underwater target recognition result annotated with the time-frequency power spectrum.

[0007] In some embodiments of this application, step S1 further includes: Electromagnetic data is captured by a sliding window of fixed duration; The time sliding window is moved at fixed time intervals, and the electromagnetic data captured by the time sliding window at all time intervals is used as electromagnetic data samples.

[0008] In some embodiments of this application, the electromagnetic detection method further includes: Electromagnetic data captured by a fixed-time long-term sliding window is used as a time-domain signal sequence and input into a time-domain signal deep learning network. Electromagnetic data captured by a fixed-time long-term sliding window is used to calculate the power spectrum using the power spectrum estimation method. The obtained frequency domain power spectrum data is used as a frequency domain power spectrum signal sequence and input into a frequency domain power spectrum signal deep learning network. Electromagnetic data captured by a fixed-time long-term sliding window is used to calculate the time-frequency power spectrum using the power spectrum estimation method and the sliding window method. The time-frequency power spectrum data obtained by the fixed-time long-term sliding window is used as the time-frequency power spectrum graph and input into the time-frequency power spectrum signal deep learning network.

[0009] In some embodiments of this application, the comprehensive decision in step S4, which generates the final underwater target identification result annotated with time-frequency power spectrum, further includes: If the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network all identify the presence of an underwater target, then it is determined to be a target of the first intensity level. If only one of the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identifies the existence of an underwater target, and the time-frequency power spectrum signal deep learning network identifies the existence of an underwater target, then it is determined to be a target identified at the second intensity level. If both the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identify that there is no underwater target, and the time-frequency power spectrum signal deep learning network identifies that there is an underwater target, then it is judged as a target of the third intensity level. If both the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identify the presence of an underwater target, and the time-frequency power spectrum signal deep learning network identifies the absence of an underwater target, then the target is identified as a fourth intensity level target. If only one of the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identifies the existence of an underwater target or both identify the absence of an underwater target, and the time-frequency power spectrum signal deep learning network identifies the absence of an underwater target, then it is determined that no target is identified. The accuracy of target recognition decreases gradually according to the order of target identification at the first intensity level, target identification at the second intensity level, target identification at the third intensity level, target identification at the fourth intensity level, and no target identified.

[0010] In some embodiments of this application, step S1 further includes: collecting underwater electromagnetic signals through electromagnetic detection nodes arranged at different locations connected sequentially via a wired network, and obtaining underwater electromagnetic data and electromagnetic data samples. The intelligent electromagnetic detection method for underwater targets also includes: integrating underwater electromagnetic data samples from various electromagnetic detection nodes to identify underwater targets and their movement trends.

[0011] In some embodiments of this application, step S1 further includes: acquiring underwater electromagnetic signals, obtaining underwater electromagnetic data, performing noise reduction processing on the underwater electromagnetic data, and using the noise-reduced underwater electromagnetic data as an electromagnetic data sample.

[0012] In some embodiments of this application, a network model training step is also included, comprising: Based on the underwater target recognition results of the time-domain signal deep learning network, the time-domain signal sequence is labeled as having underwater target time-domain signal features and not having underwater target time-domain signal features. The labeled time-domain signal sequence is then used to construct a training set for the time-series signal deep learning network to train the network. Based on the underwater target recognition results of the frequency domain power spectrum signal deep learning network, the frequency domain power spectrum signal sequence is labeled as having underwater target frequency domain power spectrum signal features and not having underwater target frequency domain power spectrum signal features. The labeled frequency domain power spectrum signal sequence is constructed as a training set for the frequency domain power spectrum signal deep learning network, which is used to train the frequency domain power spectrum signal deep learning network. Based on the underwater target identification results of the time-frequency power spectrum signal deep learning network, the time-frequency power spectrum is marked as having underwater target characteristics and not having underwater target characteristics. The electric and magnetic field characteristics of the underwater target are distinguished by the marking, and the time of the time-frequency power spectrum is marked as the stage before the underwater target appears, the stage during the underwater target appears, or the stage after the underwater target appears. The marked time-frequency power spectrum is constructed as a training set for the time-frequency power spectrum signal deep learning network, which is used to train the time-frequency power spectrum signal deep learning network.

[0013] This application also provides an intelligent electromagnetic detection system for underwater targets, capable of executing the intelligent electromagnetic detection method for underwater targets provided in the foregoing embodiments, which includes: Electromagnetic detection subsystem: includes multiple electromagnetic detection nodes, which are sequentially connected through a wired network. It is used to collect underwater electromagnetic signals and obtain electromagnetic data samples of underwater electromagnetic signals. Electromagnetic monitoring and intelligent processing subsystem: It includes: Electromagnetic monitoring unit: Communicates with the electromagnetic detection subsystem via a wired network to acquire electromagnetic data samples of underwater electromagnetic signals and stores them in the electromagnetic database; Electromagnetic data preprocessing unit: used to denoise the electromagnetic data samples of underwater electromagnetic signals acquired by the electromagnetic monitoring unit, and further decompose them into time-domain signal sequence, frequency-domain power spectrum signal sequence and time-frequency power spectrum diagram; An electromagnetic data intelligent processing unit includes: A time-domain signal deep learning network is used to take a time-domain signal sequence as input and identify the presence of underwater targets. A deep learning network for frequency domain power spectrum signals is used to take a sequence of frequency domain power spectrum signals as input and identify the presence of underwater targets. A deep learning network for time-frequency power spectrum signals is used to take time-frequency power spectrum maps as input and identify the presence of underwater targets. The comprehensive decision unit is connected to the outputs of the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network. It is configured to generate the final underwater target identification result according to the comprehensive decision rules based on the underwater target identification results of the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network.

[0014] In some embodiments of this application, each electromagnetic detection node includes: an electric field sensor, a magnetic field sensor, an electric field pre-signal processing circuit, a magnetic field pre-signal processing circuit, an acquisition circuit, and a main controller; the acquisition circuit is a multi-channel synchronous sampling analog-to-digital converter chip; the main controller includes an FPGA and an MCU; the FPGA is connected to the acquisition circuit to control multi-channel synchronous sampling; the electric field sensor is connected to the acquisition circuit via the electric field pre-signal processing circuit; the magnetic field sensor is connected to the acquisition circuit via the magnetic field pre-signal processing circuit.

[0015] In some embodiments of this application, the electric field pre-amplifier signal processing circuit includes, in sequence, an input impedance matching unit, a chopper modulation unit, an audio transformer isolation amplification unit, a high input impedance pre-amplifier unit, a low-noise integrated operational amplifier amplification unit, a programmable gain unit, a chopper demodulation unit, a low-pass filter unit, and an output impedance matching unit; the input impedance matching unit is connected to the electric field sensor, and the output impedance matching unit is connected to the acquisition circuit. The magnetic field pre-processing circuit includes an input impedance matching unit, a low-noise integrated operational amplifier amplification unit, and an output impedance matching unit. The input impedance matching unit is connected to the magnetic field sensor, and the output impedance matching unit is connected to the acquisition circuit.

[0016] Compared with existing technologies, the intelligent electromagnetic detection system and method for underwater targets provided in this application have at least the following beneficial effects: 1. This invention proposes an intelligent electromagnetic detection system and method for underwater targets, which applies artificial intelligence technology to the electromagnetic detection of underwater targets, realizes intelligent electromagnetic detection of underwater targets, improves the detection capability of underwater targets, and has good detection effect.

[0017] 2. The intelligent electromagnetic detection system and method for underwater targets proposed in this invention processes the collected electromagnetic signals into time-domain signals, frequency-domain power spectrum signals, and time-frequency power spectrum signals, and learns them through their respective independent deep learning networks. Each deep learning network can better adapt to the characteristics of various signals, thereby enabling more targeted training of the deep learning network. Furthermore, through comprehensive decision-making, more accurate underwater target identification can be achieved.

[0018] 3. The underwater target intelligent electromagnetic detection system and method proposed in this invention uses a sliding time window method with a fixed time interval of fixed duration to intercept electromagnetic data for intelligent processing, which improves the real-time performance of underwater target detection while meeting the computation time requirements of the intelligent processing algorithm.

[0019] 4. The intelligent electromagnetic detection system and method for underwater targets proposed in this invention can identify the electromagnetic signal patterns of underwater targets at different time stages, such as before, during, or after the appearance of underwater targets. Furthermore, multiple electromagnetic detection nodes are sequentially connected through a wired network to form an underwater target electromagnetic detection array, which can identify the movement trend of underwater targets while identifying them.

[0020] 5. The intelligent electromagnetic detection system and method for underwater targets proposed in this invention includes an electric field pre-amplification signal processing circuit consisting of an input impedance matching unit, a chopper modulation unit, an audio transformer isolation amplification unit, a high input impedance pre-amplification unit, a low-noise integrated operational amplifier amplification unit, a programmable gain unit, a chopper demodulation unit, a low-pass filter unit, and an output impedance matching unit connected in sequence. This circuit enables signal amplification processing of extremely low-frequency weak electric field signals at the nV level with ultra-low background noise. It employs a DC-coupled low input impedance matching method to access extremely low-frequency weak electric field signals; it uses a chopper modulation amplification method of "chopper modulation + chopper demodulation" to suppress 1 / F noise in the low-noise operational amplifier; it uses a three-stage fixed-gain amplification process of "audio transformer isolation amplification + high input impedance pre-amplification + low-noise integrated operational amplifier amplification" to achieve signal amplification processing of extremely low background noise at the nV level with channel signal isolation; and it uses programmable gain amplification to achieve a large measurement dynamic range.

[0021] 6. The underwater target intelligent electromagnetic detection system proposed in this invention can be applied to various application scenarios such as shore-based fixed electromagnetic detection, ship-borne towed electromagnetic detection, and underwater mobile electromagnetic detection, providing effective technical support for the development of underwater target electromagnetic detection technology and equipment. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the underwater target intelligent electromagnetic detection system according to an embodiment of this application; Figure 2 This is a logical structure diagram of an electromagnetic detection node according to an embodiment of this application; Figure 3 This is a digital logic block diagram of a multi-channel synchronous acquisition of an electromagnetic detection node according to an embodiment of this application; Figure 4 This is a block diagram of the electric field pre-signal processing circuit of the electromagnetic detection node according to an embodiment of this application; Figure 5 This is a block diagram of the magnetic field pre-signal processing circuit of the electromagnetic detection node according to an embodiment of this application; Figure 6 This is a system network topology diagram of the electromagnetic detection subsystem based on Ethernet according to the embodiments of this application; Figure 7 This is a block diagram of the shipborne electromagnetic monitoring and intelligent processing subsystem according to the embodiments of this application; Figure 8 This is a flowchart of the intelligent processing method for electromagnetic identification of underwater targets that combines time and frequency domains according to the embodiments of this application. Figure 9 This is a comprehensive decision logic diagram for intelligent electromagnetic detection of underwater targets according to an embodiment of this application; Figure 10 This is a flowchart illustrating the construction of a training dataset for a deep learning network according to an embodiment of this application. Figure 11 It is a time-frequency power spectrum diagram marked as the previous stage according to the embodiments of this application; Figure 12 It is a time-frequency power spectrum diagram marked as the middle stage according to the embodiments of this application; Figure 13 It is a time-frequency power spectrum marked as the later stage according to the embodiments of this application; Figure 14 This is a schematic diagram of an underwater target joint detection method of a multi-node electromagnetic detection system according to an embodiment of this application; Figure 15 The results are the background noise test results of the electric field and magnetic field channels of the electromagnetic detection node according to the embodiments of this application; Figure 16 The results are the real-time transmission performance test results of the wired network of the electromagnetic detection node according to the embodiments of this application; Figure 17 The network structure and parameters of the YOLOv5s lightweight underwater target detection network used in the embodiments of this application are described. Figure 18 The magnetic field test result at a certain moment is the result of the intelligent processing method for electromagnetic identification of underwater targets according to the embodiments of this application. Figure 19 The magnetic field test result at a certain moment is the result of the intelligent processing method for electromagnetic identification of underwater targets according to the embodiments of this application. Figure 20 The magnetic field test result at a certain moment is the result of the intelligent processing method for electromagnetic identification of underwater targets according to the embodiments of this application. Figure 21 The result is the electric field test result at a certain moment of the underwater target electromagnetic identification intelligent processing method according to the embodiments of this application; Figure 22 The result is the electric field test result at a certain moment of the underwater target electromagnetic identification intelligent processing method according to the embodiments of this application; Figure 23 The result is the electric field test result at a certain moment of the underwater target electromagnetic identification intelligent processing method according to the embodiments of this application; Figure 24 This is a schematic diagram of a shore-based fixed electromagnetic detection application according to an embodiment of this application; Figure 25 This is a schematic diagram of a shipborne towed electromagnetic detection application according to an embodiment of this application. Figure 26 This is a schematic diagram of an underwater mobile electromagnetic detection application equipped with an AUV, according to an embodiment of this application; Figure 27 This is a flowchart of an intelligent electromagnetic detection method for underwater targets according to an embodiment of this application. Detailed Implementation

[0024] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0025] The prefixes such as "first" and "second" used in this embodiment are merely for distinguishing different descriptive objects and do not limit the position, order, priority, quantity, or content of the described objects. The use of ordinal numbers and other prefixes to distinguish descriptive objects in this embodiment does not constitute a limitation on the described objects. The description of the described objects is given in the context of the embodiments, and the use of such prefixes should not constitute unnecessary restrictions. Furthermore, in the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0026] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B; the term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone.

[0027] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0028] This application proposes an intelligent electromagnetic detection system for underwater targets, the composition of which is referenced in the embodiments. Figure 1 It includes an electromagnetic detection subsystem and an electromagnetic monitoring and intelligent processing subsystem.

[0029] The electromagnetic detection subsystem comprises multiple electromagnetic detection nodes for acquiring underwater electromagnetic signals. These nodes are connected sequentially, and each node includes an electric field sensor, a magnetic field sensor, a pre-processing circuit, a main controller, and an acquisition circuit. For example, the electric field sensor uses Ag / AgCl electrodes or carbon fiber electrodes suitable for marine environments; the magnetic field sensor uses a triaxial fluxgate sensor.

[0030] Specifically, the electromagnetic detection nodes are spaced apart underwater, capable of collecting electromagnetic signals from different locations. The nodes are connected via a wired network, and the logic structure diagram of each node is shown in the reference diagram. Figure 2 The sensing unit includes an electric field sensor and a magnetic field sensor; the main controller includes an FPGA and an MCU. The MCU connects to an SD card and a network communication interface, and is responsible for system management, data storage, and data transmission. The FPGA connects to two 24-bit, 4-channel synchronous sampling analog-to-digital converters (ADCs), which are connected to the electric field sensor and the magnetic field sensor respectively, enabling synchronous acquisition of 24-bit, 8-channel signals. A high-precision clock provides the MCU and FPGA with a high-precision time base signal for high-precision synchronous acquisition, data storage, and data transmission.

[0031] To ensure the accuracy of underwater target identification, the synchronization of acquired electromagnetic signals is crucial. Therefore, this application proposes a synchronous acquisition logic architecture and method for electromagnetic detection nodes.

[0032] refer to Figure 3 In this embodiment, synchronous signal acquisition is implemented using an FPGA. The digital logic of the FPGA includes: 4-channel synchronous sampling ADC sampling control logic unit: The 4-channel synchronous sampling ADC sampling control logic function includes the interface signal timing, register configuration, sampling status timing, etc. of the 4-channel synchronous sampling ADC, to realize the acquisition of data by the 4-channel synchronous sampling ADC; Channel data extraction logic unit: used for data extraction from each channel, realizing channel data separation; Data time stamping logic unit: Adds a high-precision time stamp generated by RTC clock logic to a set of data for each channel according to the data storage format; Ping-pong RAM data buffer unit: used for continuous ADC sampling of each channel; Data buffer interrupt unit: used to generate an interrupt signal to notify the MCU to read the buffer data when all 8 channels of ping-pong RAM data buffer are full; RTC clock logic unit: Used to provide a high-precision clock for the sampling control logic and data timing logic of the 4-channel synchronous sampling ADC, so as to achieve 8-channel high-precision synchronous acquisition.

[0033] To achieve high-precision detection of electromagnetic signals, a pre-processing circuit is used to pre-process the acquired electromagnetic signals. The processed electromagnetic signals are then input to a 4-channel synchronous sampling ADC. Due to the different characteristics of electrical and magnetic signals, this embodiment further includes an electric field pre-processing circuit and a magnetic field pre-processing circuit.

[0034] The electric field preamplifier signal processing circuit includes, in sequence, an input impedance matching unit, a chopper modulation unit, an audio transformer isolation amplification unit, a high input impedance preamplifier unit, a low-noise integrated operational amplifier amplification unit, a programmable gain unit, a chopper demodulation unit, a low-pass filter unit, and an output impedance matching unit. The input impedance matching unit is connected to the electric field sensor, and the output impedance matching unit is connected to the 4-channel synchronous sampling ADC of the main control unit. The magnetic field preamplifier signal processing circuit includes an input impedance matching unit, a low-noise integrated operational amplifier amplification unit, and an output impedance matching unit. The input impedance matching unit is connected to the magnetic field sensor, and the output impedance matching unit is connected to the 4-channel synchronous sampling ADC of the main control unit.

[0035] refer to Figure 4This circuit is one of the core technologies for weak electric field signal detection and processing, achieving signal amplification with ultra-low noise floor (nV level) for extremely low-frequency weak electric field signals. It employs a DC-coupled low-input impedance matching method to access the extremely low-frequency weak electric field signal; a chopper modulation amplification method combining chopper modulation and demodulation to suppress 1 / F noise in the low-noise operational amplifier; a three-stage fixed-gain amplification system consisting of an audio transformer for isolation, a high-input-impedance preamplifier, and a low-noise integrated operational amplifier to achieve ultra-low noise floor (nV level) signal amplification with channel signal isolation; and a programmable gain amplification to achieve a large measurement dynamic range. A low-pass filter at the circuit's rear is used to remove chopper modulation noise, and output impedance matching is used to achieve a low-resistance amplified signal output.

[0036] refer to Figure 5 This circuit achieves low-noise amplification of the output signal of the magnetic field sensor. It is implemented using a low-noise integrated operational amplifier and consists of three parts: input impedance matching, low-noise integrated operational amplifier amplification, and output impedance matching. Input impedance matching enables the input of magnetic field signals with high input impedance, low-noise integrated operational amplifier amplification amplifies the magnetic field signal, and output impedance matching enables the output of amplified magnetic field signals with low output resistance.

[0037] In some embodiments of this application, the electromagnetic detection nodes are connected via a wired network. Each electromagnetic detection node can be connected to the electromagnetic monitoring and intelligent processing subsystem separately, or it can ultimately be connected to the electromagnetic monitoring and intelligent processing subsystem via a main electromagnetic detection node.

[0038] like Figure 6 As shown, the wired network uses Ethernet. Each electromagnetic detection node is configured with a network communication interface. In this embodiment, each electromagnetic detection node is configured with a three-port Ethernet router. Each three-port Ethernet router includes one WAN interface and two LAN interfaces (LAN1 and LAN2). The LAN1 interface of the current electromagnetic detection node is connected to its electromagnetic detection circuit, and the LAN2 interface of the current electromagnetic detection node is connected to the WAN interface of the backend electromagnetic detection node.

[0039] The underwater electromagnetic data collected by the electromagnetic detection subsystem contains underwater target characteristics. The electromagnetic data is transmitted to the electromagnetic monitoring and intelligent processing subsystem, which then completes the underwater target signal identification.

[0040] Structural Reference of Electromagnetic Monitoring and Intelligent Processing Subsystem Figure 7 It connects to the electromagnetic detection subsystem via a network communication interface, and includes: Electromagnetic monitoring unit: It is configured with a network communication interface and connects to an electromagnetic detection node through the network communication interface. Since the electromagnetic detection nodes communicate in series, the electromagnetic monitoring unit can acquire the underwater electromagnetic signals collected by each electromagnetic detection node and store the underwater electromagnetic data received through the network communication interface in the electromagnetic database. An intelligent processing unit for underwater target identification acquires and processes electromagnetic data from an electromagnetic database, including: Electromagnetic data preprocessing unit: used to denoise the underwater electromagnetic data acquired by the electromagnetic monitoring unit and further decompose it into time domain signal sequence, frequency domain power spectrum signal sequence and time-frequency power spectrum diagram; An electromagnetic data intelligent processing unit includes: A time-domain signal deep learning network is used to take a time-domain signal sequence as input and identify the presence of underwater targets. A deep learning network for frequency domain power spectrum signals is used to take a sequence of frequency domain power spectrum signals as input and identify the presence of underwater targets. A deep learning network for time-frequency power spectrum signals is used to take time-frequency power spectrum maps as input and identify the presence of underwater targets. The comprehensive decision unit is connected to the outputs of the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network. It is configured to generate the final underwater target identification result according to the comprehensive decision rules based on the underwater target identification results of the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network.

[0041] In some embodiments of this application, the electromagnetic monitoring unit includes an output storage module and a data display module. The data storage module uses a data memory to store the acquired electromagnetic data; the data display module uses a data display to show the real-time electromagnetic data of each electromagnetic detection node, and also displays the underwater target identification results based on the output information of the intelligent electromagnetic data processing method. The intelligent underwater target identification processing unit acquires electromagnetic data from the electromagnetic database, runs the electromagnetic data preprocessing method and the intelligent underwater target identification processing method, and achieves underwater target identification.

[0042] It should be noted that the electromagnetic data of each electromagnetic detection node in the underwater target intelligent electromagnetic detection system is processed independently. After the electromagnetic data of each electromagnetic detection node is acquired and preprocessed, the time-domain signal sequence, frequency-domain power spectrum signal sequence, and time-frequency power spectrum graph are used as network inputs respectively. Three independent deep learning networks are used to identify underwater target information. After comprehensively judging the identification results of the three deep learning networks, the underwater target identification result of each electromagnetic node marked with a time-frequency power spectrum graph is finally given.

[0043] It should be understood that the three deep learning network models can be the same or different, and they can be either heavyweight or lightweight networks. For example, depending on the computing platform resources, the deep learning networks can be either typical heavyweight networks such as LeNet, AlexNet, VGG, ResNet, and GoogleNet, or lightweight networks such as YOLO, SqueezeNet, MobileNet, and ShuffleNet. Figure 17 An example of a deep learning neural network model is given.

[0044] Based on the above-described intelligent electromagnetic detection system for underwater targets, this application provides a further intelligent electromagnetic detection method for underwater targets.

[0045] S1: Collect underwater electromagnetic signals and obtain electromagnetic data samples of the underwater electromagnetic signals.

[0046] Specifically, this application acquires underwater electromagnetic signals and obtains underwater electromagnetic data through an electromagnetic detection subsystem. After acquiring the underwater electromagnetic data, the electromagnetic detection subsystem transmits it to the electromagnetic monitoring and intelligent processing subsystem, whereby the electromagnetic monitoring unit stores the underwater electromagnetic data in an electromagnetic database.

[0047] During the process of obtaining electromagnetic data samples, the collected electromagnetic data is captured by a fixed-duration sliding window. A time sliding window of fixed duration is moved at fixed time intervals, and the electromagnetic data captured by the time sliding window at all time intervals is used as electromagnetic data samples.

[0048] Specifically, to improve the real-time performance of underwater target electromagnetic identification, this invention employs a sliding time window method to acquire electromagnetic data input to the intelligent processing network each time. Electromagnetic data is captured within a time window of a certain duration (e.g., 2 hours), and the time window slides at regular intervals (e.g., 5 minutes), allowing for real-time acquisition of electromagnetic data to be processed from the electromagnetic database. The sliding time window method (commonly known as the sliding window method) is a data processing technique that analyzes local features by moving a fixed-size window across a data sequence. This method maintains a fixed-size window, which slides progressively across the data sequence, moving one or more units at a time. The window contains the currently processed data points, and statistical quantities (such as sums and maximum values) or detection patterns are efficiently calculated by dynamically adjusting the start and end positions of the window.

[0049] To improve recognition accuracy, in step S1, after acquiring underwater electromagnetic data, further denoising processing is performed on the underwater electromagnetic data, and the denoised underwater electromagnetic data is used as a data sample. Denoising processing methods include, but are not limited to, filtering and normalization.

[0050] S2: Decompose the electromagnetic data sample into a time-domain signal sequence, a frequency-domain power spectrum signal sequence, and a time-frequency power spectrum.

[0051] S3: Input the time-domain signal sequence into the time-domain signal deep learning network to identify whether there is an underwater target in the time-domain signal sequence; input the frequency-domain power spectrum signal sequence into the frequency-domain power spectrum signal deep learning network to identify whether there is an underwater target in the frequency-domain power spectrum signal sequence; input the time-frequency power spectrum graph into the time-frequency power spectrum signal deep learning network to identify whether there is an underwater target in the time-frequency power spectrum graph.

[0052] In some embodiments of this application, after decomposing the electromagnetic data samples into a time-domain signal sequence, a frequency-domain power spectrum signal sequence, and a time-frequency power spectrum graph signal, the electromagnetic data captured by a fixed-length sliding window is used as a time-domain signal sequence and input into a time-domain signal deep learning network; the electromagnetic data captured by a fixed-length sliding window is used to perform power spectrum calculation using a power spectrum estimation method, and the obtained frequency-domain power spectrum data is used as a frequency-domain power spectrum signal sequence and input into a frequency-domain power spectrum signal deep learning network; the electromagnetic data captured by a fixed-length sliding window is used to perform time-frequency power spectrum calculation using a power spectrum estimation method and a sliding window method, and the obtained time-frequency power spectrum data of the fixed-length sliding window is used as a time-frequency power spectrum graph and input into a time-frequency power spectrum signal deep learning network.

[0053] For example, the preprocessed electromagnetic data is used as network input data in three signal forms: time-domain signal sequence, frequency-domain power spectrum signal sequence, and time-frequency power spectrum. The time-domain signal sequence is electromagnetic data captured according to a certain time window (e.g., 2 hours) and directly input to the intelligent processing network. The frequency-domain power spectrum signal sequence is electromagnetic data captured according to a certain time window (e.g., 2 hours) and converted into frequency-domain power spectrum data using the Welch method and then input to the intelligent processing network. The time-frequency power spectrum is electromagnetic data captured according to a certain time window (e.g., 2 hours) and converted into a time-frequency power spectrum using the "Welch method + sliding time window method" and then input to the intelligent processing network. The time window length for calculating the time-frequency power spectrum is 128 points, a 50% overlap rate is used to reduce spectral leakage and improve spectral resolution, and a Hanning window function is used to reduce edge effects and improve smoothness.

[0054] S4: Based on the underwater target identification results from the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network, the final underwater target identification result is generated. The underwater target identification result with time-spectrum annotation is given based on the comprehensive decision result.

[0055] In some embodiments of this application, based on the characteristics of different signals, the network outputs of two deep learning networks—one for a time-domain signal sequence and the other for a frequency-domain power spectrum signal sequence—are binarized outputs with and without targets; for example, a target is labeled as 1, and no target is labeled as 0. The network output of the deep learning network for the time-frequency power spectrum map is the underwater target identification output labeled on the time-frequency power spectrum map.

[0056] In some embodiments of this application, the underwater target identification results of various learning networks are combined to generate a final underwater target identification result, specifically including targets identified at a first intensity level, targets identified at a second intensity level, targets identified at a third intensity level, targets identified at a fourth intensity level, and no targets identified. The order of targets identified at the first intensity level, the second intensity level, the third intensity level, the fourth intensity level, and no targets identified indicates that the accuracy of target identification gradually decreases. (Reference) Figure 9 The specific identification method is as follows.

[0057] If the time-domain signal deep learning network, the frequency domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network all identify the presence of an underwater target, then it is judged as a target with the first intensity level. If only one of the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identifies the existence of an underwater target, and the time-frequency power spectrum signal deep learning network also identifies the existence of an underwater target, then it is judged as a target identified at the second intensity level. If both the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identify that there is no underwater target, but the time-frequency power spectrum signal deep learning network identifies that there is an underwater target, then it is judged as a target of the third intensity level. If both the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identify the presence of an underwater target, but the power spectrum signal learning network does not identify the presence of an underwater target, then it is judged as a target identified at the fourth intensity level. If only one of the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identifies the existence of an underwater target, or both identify the absence of an underwater target, and the time-frequency power spectrum signal deep learning network identifies the absence of an underwater target, then it is determined that no target is identified.

[0058] In some embodiments of this application, the decision result is categorized into no target, target identification at the fourth intensity level, target identification at the third intensity level, target identification at the second intensity level, and target identification at the first intensity level. When the decision result is a target identification at the first, second, or third intensity level, the deep learning network using the time-frequency power spectrum as its input can output a network output result with a target. The target signal morphology identified by the time-frequency power spectrum can be labeled as three different target appearance time stages: before the target, during the target, and after the target. Specifically, "before the target" indicates that the current data frame is in the first half of the underwater target features; "during the target" indicates that the underwater target features are fully present in the current data frame; and "after the target" indicates that the current data frame is in the second half of the underwater target features.

[0059] Specifically, the deep learning network training for the time-domain and frequency-domain joint intelligent processing method for underwater target electromagnetic identification based on deep learning networks in this application embodiment requires a dataset. The dataset construction method is as follows: Figure 10 As shown, the electromagnetic data acquisition and preprocessing methods are the same as those in the intelligent processing workflow. The training datasets for the time-domain signal sequence and the frequency-domain power spectrum signal sequence are labeled as having / not having underwater targets based on the time period information of the acquired electromagnetic data and the presence or absence of underwater targets within that time period. The training dataset for the time-frequency power spectrum is further labeled as electric and magnetic field characteristic signals based on the generated time-frequency power spectrum, and divided into three stages—pre-stage, mid-stage, and post-stage—according to the signal characteristic morphology. An example of the labeling of the training set for the time-frequency power spectrum is shown below. Figures 11 to 13 As shown.

[0060] In some embodiments of this application, to obtain more comprehensive electromagnetic detection data, the electromagnetic detection subsystem includes multiple electromagnetic detection nodes distributed at different underwater locations. These nodes can collect electromagnetic detection signals from different underwater locations. Since underwater targets are typically moving, the movement of underwater targets can be reflected based on electromagnetic detection signals from different underwater locations and at different times. Therefore, step S1 further includes: collecting underwater electromagnetic signals from the electromagnetic detection nodes arranged at different locations sequentially connected via a wired network to obtain underwater electromagnetic data and underwater electromagnetic data samples; step S4 further includes: combining the underwater electromagnetic data samples obtained from the underwater electromagnetic data collected by each electromagnetic detection node to identify underwater targets and their movement trends.

[0061] The underwater target intelligent electromagnetic detection system and method provided in the embodiments of this application were tested, and the test results are as follows.

[0062] (1) Test results of background noise of electric and magnetic field channels The background noise test results of the electric and magnetic field channels of the electromagnetic detection node are as follows: Figure 15 As shown in the figure, Ex, Ey, and Ez represent three electric field channels, processed using an electric field pre-processing circuit; Mx, My, Mz, Hx, and Hy represent five magnetic field channels, processed using a magnetic field pre-processing circuit. It can be seen that the noise floor of the electric field channels is 1.585~3 nV / rt(Hz)@1Hz; the noise floor of the magnetic field channels is 1.579 uV / rt(Hz)@1Hz. Connecting to a fluxgate sensor with a sensitivity of 100 mV / uT, the noise floor of the magnetic field channels, converted to magnetic field strength, is 15.79 pT / rt(Hz)@1Hz, which is comparable to the noise level of a fluxgate sensor with a noise floor of 10~20 pT / rt(Hz)@1Hz. This meets the requirements for acquiring nV-level electric field and pT-level magnetic field signals from underwater targets.

[0063] (2) Real-time transmission performance test results of wired networks The real-time transmission performance of wired networks is essential for ensuring the real-time detection capability of multi-electromagnetic detection node underwater target electromagnetic detection systems. Figure 16 The results show the network transmission capability test results using the STM32H7 series high-performance MCU. It can be seen that the network transmission rate is between 4MB / s and 7MB / s, with an average network transmission rate of 4.9MB / s. Based on a network transmission efficiency of 40%, it can theoretically meet the data transmission requirements of approximately 20 8-channel electromagnetic detection nodes with 500Hz sampling.

[0064] (3) Test results of the intelligent processing method for electromagnetic identification of underwater targets Figure 17The diagram shows the network structure and parameters of the lightweight underwater target detection network YOLOv5s. In the diagram, Input is the network input, Backbone performs feature extraction, Neck is responsible for fusing shallow features from the extracted feature maps at three scales, and Head performs target prediction. Specifically, Conv is the convolutional layer module, C3_1, C3_2, and C3_3 are modules containing three convolutional layers and one bottleneck layer, SPPF is the spatial pyramid pooling module, Upsample is the upsampling module, Concat is the feature fusion module, and Detect is the target detection module. Figures 18 to 23 The test results at different times are for time-frequency spectrum annotation and recognition of underwater target electromagnetic recognition intelligent processing method using YOLOv5s lightweight target detection network. Figures 18 to 23 C1, C2, and C3 represent the three stages of the underwater target's magnetic field signal: the early, middle, and late stages. Figures 18 to 23 In the diagram, d1, d2, and d3 represent the three stages of the underwater target electric field signal: the beginning, the middle, and the end. It can be seen that the method of this invention using YOLOv5s can clearly distinguish the time-spectrum information of the underwater target electric and magnetic fields at different stages, demonstrating good intelligent recognition capabilities.

[0065] The underwater target intelligent electromagnetic detection system provided in this application is mainly aimed at shore-based fixed electromagnetic detection and ship-borne towed electromagnetic detection application scenarios based on wired networks. Its simplified single electromagnetic detection node system can be extended to mobile electromagnetic detection application scenarios based on underwater mobile platforms.

[0066] (1) Application scenarios of shore-based fixed electromagnetic detection Figure 24 The diagram shows an application scenario of the shore-based fixed electromagnetic detection system of the present invention. In this application, the system includes an electromagnetic monitoring and intelligent processing subsystem deployed on the shore and an electromagnetic detection subsystem deployed underwater near the shore.

[0067] (2) Application scenarios of shipborne towed electromagnetic detection Figure 25 The diagram shows a shipborne towed electromagnetic detection application scenario of the present invention. In this application, the system includes an electromagnetic monitoring and intelligent processing subsystem deployed on a ship and an electromagnetic detection subsystem towed underwater.

[0068] (3) Electromagnetic detection application scenarios equipped with underwater mobile platforms such as AUVs Figure 26 This is a simplified diagram illustrating the application of the present invention as a single electromagnetic node in an electromagnetic detection scenario mounted on an underwater mobile platform such as an AUV. In this application, both the electromagnetic monitoring and intelligent processing subsystem and the electromagnetic detection node are deployed inside the underwater mobile platform.

[0069] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for intelligent electromagnetic detection of underwater targets, characterized in that, Includes the following steps: S1: Collect underwater electromagnetic signals to obtain underwater electromagnetic data and electromagnetic data samples; S2: Decompose the electromagnetic data sample into a time-domain signal sequence, a frequency-domain power spectrum signal sequence, and a time-frequency power spectrum diagram; S3: Input the time-domain signal sequence into a time-domain signal deep learning network to identify whether there is an underwater target in the time-domain signal sequence; input the frequency-domain power spectrum signal sequence into a frequency-domain power spectrum signal deep learning network to identify whether there is an underwater target in the frequency-domain power spectrum signal sequence; input the time-frequency power spectrum graph into a time-frequency power spectrum signal deep learning network to identify whether there is an underwater target in the time-frequency power spectrum graph. S4: Based on the underwater target recognition results of the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network, a comprehensive decision is made to generate the final underwater target recognition result annotated with the time-frequency power spectrum.

2. The intelligent electromagnetic detection method for underwater targets according to claim 1, characterized in that, Step S1 also includes: Electromagnetic data is captured by a sliding window of fixed duration; The time sliding window is moved at fixed time intervals, and the electromagnetic data captured by the time sliding window at all time intervals is used as electromagnetic data samples.

3. The intelligent electromagnetic detection method for underwater targets according to claim 2, characterized in that, Also includes: Electromagnetic data captured by a fixed-time long-term sliding window is used as a time-domain signal sequence and input into a time-domain signal deep learning network. Electromagnetic data captured by a fixed-time long-term sliding window is used to calculate the power spectrum using the power spectrum estimation method. The obtained frequency domain power spectrum data is used as a frequency domain power spectrum signal sequence and input into a frequency domain power spectrum signal deep learning network. Electromagnetic data captured by a fixed-time long-term sliding window is used to calculate the time-frequency power spectrum using the power spectrum estimation method and the sliding window method. The time-frequency power spectrum data obtained by the fixed-time long-term sliding window is used as the time-frequency power spectrum graph and input into the time-frequency power spectrum signal deep learning network.

4. The intelligent electromagnetic detection method for underwater targets according to claim 1, characterized in that, Step S4, the comprehensive decision that generates the final underwater target identification result with time-frequency power spectrum annotation, also includes: If the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network all identify the presence of an underwater target, then it is determined to be a target of the first intensity level. If only one of the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identifies the existence of an underwater target, and the time-frequency power spectrum signal deep learning network identifies the existence of an underwater target, then it is determined to be a target identified at the second intensity level. If both the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identify that there is no underwater target, and the time-frequency power spectrum signal deep learning network identifies that there is an underwater target, then it is judged as a target of the third intensity level. If both the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identify the presence of an underwater target, and the time-frequency power spectrum signal deep learning network identifies the absence of an underwater target, then the target is identified as a fourth intensity level target. If only one of the time-domain signal deep learning network and the frequency-domain power spectrum signal deep learning network identifies the existence of an underwater target or both identify the absence of an underwater target, and the time-frequency power spectrum signal deep learning network identifies the absence of an underwater target, then it is determined that no target is identified. The accuracy of target recognition decreases gradually according to the order of target identification at the first intensity level, target identification at the second intensity level, target identification at the third intensity level, target identification at the fourth intensity level, and no target identified.

5. The intelligent electromagnetic detection method for underwater targets according to claim 1, characterized in that: Step S1 further includes: collecting underwater electromagnetic signals through electromagnetic detection nodes arranged at different locations connected sequentially via a wired network, and obtaining underwater electromagnetic data and electromagnetic data samples. The intelligent electromagnetic detection method for underwater targets also includes: integrating underwater electromagnetic data samples from various electromagnetic detection nodes to identify underwater targets and their movement trends.

6. The intelligent electromagnetic detection method for underwater targets according to claim 1, characterized in that, Step S1 further includes: acquiring underwater electromagnetic signals, obtaining underwater electromagnetic data, performing noise reduction processing on the underwater electromagnetic data, and using the noise-reduced underwater electromagnetic data as an electromagnetic data sample.

7. The intelligent electromagnetic detection method for underwater targets according to claim 1, characterized in that, It also includes the network model training steps, including: Based on the underwater target recognition results of the time-domain signal deep learning network, the time-domain signal sequence is labeled as having underwater target time-domain signal features and not having underwater target time-domain signal features. The labeled time-domain signal sequence is then used to construct a training set for the time-series signal deep learning network to train the network. Based on the underwater target recognition results of the frequency domain power spectrum signal deep learning network, the frequency domain power spectrum signal sequence is labeled as having underwater target frequency domain power spectrum signal features and not having underwater target frequency domain power spectrum signal features. The labeled frequency domain power spectrum signal sequence is constructed as a training set for the frequency domain power spectrum signal deep learning network, which is used to train the frequency domain power spectrum signal deep learning network. Based on the underwater target identification results of the time-frequency power spectrum signal deep learning network, the time-frequency power spectrum is marked as having underwater target characteristics and not having underwater target characteristics. The electric and magnetic field characteristics of the underwater target are distinguished by the marking, and the time of the time-frequency power spectrum is marked as the stage before the underwater target appears, the stage during the underwater target appears, or the stage after the underwater target appears. The marked time-frequency power spectrum is constructed as a training set for the time-frequency power spectrum signal deep learning network, which is used to train the time-frequency power spectrum signal deep learning network.

8. An intelligent electromagnetic detection system for underwater targets, characterized in that, The intelligent electromagnetic detection method for underwater targets according to any one of claims 1 to 7 comprises: Electromagnetic detection subsystem: includes multiple electromagnetic detection nodes, which are sequentially connected through a wired network. It is used to collect underwater electromagnetic signals and obtain electromagnetic data samples of underwater electromagnetic signals. Electromagnetic monitoring and intelligent processing subsystem: It includes: Electromagnetic monitoring unit: Communicates with the electromagnetic detection subsystem via a wired network to acquire electromagnetic data samples of underwater electromagnetic signals and stores them in the electromagnetic database; Electromagnetic data preprocessing unit: used to denoise the electromagnetic data samples of underwater electromagnetic signals acquired by the electromagnetic monitoring unit, and further decompose them into time-domain signal sequence, frequency-domain power spectrum signal sequence and time-frequency power spectrum diagram; Electromagnetic data intelligent processing unit, comprising: A time-domain signal deep learning network is used to take a time-domain signal sequence as input and identify the presence of underwater targets. A deep learning network for frequency domain power spectrum signals is used to take a sequence of frequency domain power spectrum signals as input and identify the presence of underwater targets. A deep learning network for time-frequency power spectrum signals is used to take time-frequency power spectrum maps as input and identify the presence of underwater targets. The comprehensive decision unit is connected to the outputs of the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network. It is configured to generate the final underwater target identification result according to the comprehensive decision rules based on the underwater target identification results of the time-domain signal deep learning network, the frequency-domain power spectrum signal deep learning network, and the time-frequency power spectrum signal deep learning network.

9. The intelligent electromagnetic detection system for underwater targets according to claim 8, characterized in that, Each electromagnetic detection node includes: an electric field sensor, a magnetic field sensor, an electric field pre-processing circuit, a magnetic field pre-processing circuit, an acquisition circuit, and a main controller; the acquisition circuit is a multi-channel synchronous sampling analog-to-digital converter chip; the main controller includes an FPGA and an MCU; the FPGA is connected to the acquisition circuit to control multi-channel synchronous sampling; the electric field sensor is connected to the acquisition circuit via the electric field pre-processing circuit; the magnetic field sensor is connected to the acquisition circuit via the magnetic field pre-processing circuit.

10. The intelligent electromagnetic detection system for underwater targets according to claim 9, characterized in that: The electric field preamplifier signal processing circuit includes, in sequence, an input impedance matching unit, a chopper modulation unit, an audio transformer isolation amplification unit, a high input impedance preamplifier unit, a low-noise integrated operational amplifier amplification unit, a programmable gain unit, a chopper demodulation unit, a low-pass filter unit, and an output impedance matching unit; the input impedance matching unit is connected to the electric field sensor, and the output impedance matching unit is connected to the acquisition circuit. The magnetic field pre-processing circuit includes an input impedance matching unit, a low-noise integrated operational amplifier amplification unit, and an output impedance matching unit. The input impedance matching unit is connected to the magnetic field sensor, and the output impedance matching unit is connected to the acquisition circuit.

Images