Underwater online intelligent monitoring analysis system and method

Abstract

The invention belongs to the technical field of underwater sea condition data monitoring, and discloses an underwater online intelligent monitoring analysis system and method. According to the method, an acoustic Doppler current profiler (ADCP) carried by a seabed observation system is used for carrying out seabed flow measurement, the speed and direction of water flow are obtained, and a binocular camera carried by the seabed observation system is used for determining the size of underwater organisms; performing image mode recognition and classification on the underwater creatures by using an image recognition algorithm in the underwater observation system, and recognizing the types and the number of the underwater creatures; transmitting data to a communication module of the virtual anchor system unmanned platform by using the armored cable, performing error control on the data transmitted by the armored cable, and detecting and correcting error information of the data in the transmission process; and judging whether the returned information is successful or not through an information transmission control method in the underwater observation system. According to the invention, the method can achieve the image recognition of Branchiostoma under a complex sea condition, and improves the operation efficiency and safety.

Description

Technical Field

[0001] The present invention belongs to the technical field of underwater sea condition data monitoring, and in particular relates to an underwater online intelligent monitoring and analysis system and method. Background Art

[0002] The data transmission of traditional underwater image recognition systems cannot meet the requirements of complex sea conditions; the energy consumption of underwater online intelligent monitoring and analysis systems is too high and the power supply is discontinuous; the data of traditional underwater image recognition systems can only recognize images and cannot conduct comprehensive observations of the underwater environment.

[0003] Existing underwater monitoring and analysis systems can analyze images of seafloor organisms, but they lack reliable data transmission and a lack of stable energy supply, often leading to signal interruptions and information loss during operation. Furthermore, existing underwater monitoring and analysis systems use different carriers and lack comprehensive observation capabilities.

[0004] Through the above analysis, the problems and defects of the existing technology are as follows:

[0005] The existing underwater online intelligent monitoring and analysis system has distorted data transmission, unsustainable power supply and high energy consumption under complex sea conditions. Summary of the Invention

[0006] To overcome the problems existing in the related art, the embodiments disclosed in the present invention provide an underwater online intelligent monitoring and analysis system and method.

[0007] The technical solution is as follows: an underwater online intelligent monitoring and analysis method, comprising the following steps:

[0008] S1: Use the acoustic Doppler current profiler (ADCP) onboard the seabed observation system to measure seabed currents and obtain the speed and direction of the water flow. Use the binocular camera onboard the seabed observation system to determine the size of underwater organisms. Use the image recognition algorithm within the underwater observation system to perform image pattern recognition and classification on underwater organisms to identify the species and number of underwater organisms.

[0009] S2, based on all acquired data and image information, uses the armored cable to transmit data to the communication module of the virtual anchor unmanned platform, and performs error control on the data transmitted by the armored cable to detect and correct errors in the data transmission process;

[0010] S3, when the communication module transmits information back to the land base station, the information transmission control method inside the underwater observation system is used to determine whether the returned information is successful. If successful, the information is returned; otherwise, steps S1-S2 are repeated until the information is successfully returned.

[0011] In step S1, the acoustic Doppler current profiler (ADCP) measures seabed currents by emitting sound waves and receiving reflected signals. When the sound waves propagate through the water, the Doppler frequency shift caused by the water flow is measured, and the speed and direction of the water flow are calculated. The acoustic Doppler current profiler (ADCP) has multiple sound wave transmitters and uses beamforming and signal coherence processing algorithms to measure flow velocities at different depths and provide flow profile data.

[0012] Furthermore, the beamforming and signal coherence processing algorithm includes:

[0013] S101, data acquisition and preprocessing;

[0014] Acoustic wave transmission and reception: ADCP transmits multiple fixed-frequency acoustic beams into the water, receives the reflected signals and records the Doppler shift;

[0015] Frequency shift extraction, using fast Fourier transform FFT or adaptive filtering algorithm to extract frequency shift;

[0016] Time-depth stratification: stratify the echo signal by time window, combine it with the sound velocity profile, and collect the emission frequency and beam tilt angle of each depth unit;

[0017] S102, radial velocity calculation;

[0018] The radial velocity calculation formula is:

[0019]

[0020] Where, v r is the radial velocity, Δf is the Doppler shift, c is the speed of sound, f0 is the transmitting frequency, and θ is the beam tilt angle;

[0021] S103, three-dimensional flow velocity solution;

[0022] The velocity components u, v, w are solved by the improved error-weighted least squares method;

[0023] The improved error-weighted least squares method considers the difference in signal-to-noise ratio (SNR) of each beam, introduces the weight matrix W, and optimizes the solution of the velocity component. The expression is:

[0024]

[0025] Where A is the beam pointing matrix, V r is the radial velocity vector, * T is the transposed matrix;

[0026] S104, direction calculation and post-processing;

[0027] Flow velocity direction synthesis:

[0028] The horizontal velocity vector of the eastward velocity u and the northward velocity v is the direction angle for:

[0029]

[0030] Where, is the geomagnetic declination compensation value.

[0031] In step S1, the size of underwater organisms is determined using a binocular camera carried by a seabed observation system, including:

[0032] (1) Calibrate the binocular camera to determine the internal and external parameters of the camera;

[0033] (2) By acquiring the left and right images, performing feature extraction, feature matching, and disparity calculation, the depth information of the object can be perceived. The improved stereo matching algorithm is used to perceive the depth information of the object, specifically including:

[0034] Step 1: Feature extraction: Based on the deep learning feature extraction network, the feature point coordinates and descriptors are output. The expression is:

[0035]

[0036] Where F(I) is the feature extraction image, x i is the horizontal coordinate of the feature point, y i is the ordinate of the feature point, is the feature point pixel, d i For high-dimensional descriptors, cross-view consistency is optimized through contrastive learning;

[0037] Step 2: Use graph neural network-based matching optimization to perform feature matching;

[0038] Matching optimization based on graph neural networks includes: building a graph structure of feature points of the left and right images, with nodes being feature points and edges being descriptor similarities;

[0039] The graph attention network GAT aggregates neighborhood information and optimizes the matching score, which is expressed as:

[0040]

[0041] Where s ij To optimize the matching score, GAT() is an aggregation neighborhood function. is the i-th high-dimensional descriptor of the left neighborhood, is the jth high-dimensional descriptor of the right neighborhood;

[0042] Use differentiable soft matching SoftMatch to generate a probability matching matrix and combine it with polar constraints to filter out inliers;

[0043] Step 3: Disparity calculation using improved cost aggregation and optimization;

[0044] Improved cost aggregation and optimization include:

[0045] Multi-scale cost volume construction, generating multi-scale feature maps through CNN and constructing 3D cost volume C(d,x,y);

[0046] Dynamic path weighted SGM is adopted, and edge-aware weight is introduced into the smoothing term of SGM, which is expressed as:

[0047]

[0048] Where E(d) is the smoothing term of SGM, λ is the edge perception weight, p and q are the p-th and q-th paths respectively, N is the number of paths, and w pq is the edge-aware eigenvalue of the pth and qth paths, d p is the high-dimensional descriptor of the p-th path, d q is the high-dimensional descriptor of the qth path; w pq =exp(-γ||I p -I q || 2 ) is adaptively adjusted according to the image gradient, γ is the focusing factor, exp() is the image gradient adaptive adjustment function, I p is the feature point of the p-th path, I q is the characteristic point of the qth path;

[0049] Sub-pixel precision optimization, using parabola interpolation or Newton iteration to optimize parallax:

[0050]

[0051] Where, d subpixel is the sub-pixel accuracy optimization value, C is the multi-scale cost, and d is a high-dimensional descriptor;

[0052] (3) Reconstruct the three-dimensional scene based on the depth information.

[0053] In step S1, the image recognition algorithm within the underwater observation system is used to perform image pattern recognition and classification of underwater organisms, including:

[0054] First, the captured images are preprocessed with denoising and contrast enhancement;

[0055] Then, the image features are extracted through the convolutional neural network CNN;

[0056] Next, the classifier introduces an attention mechanism and optimizes the loss function to analyze the features and identify the types and numbers of underwater creatures. The attention mechanism uses an improved hybrid attention module (HAM), which includes channel attention, spatial attention, and feature fusion.

[0057] The channel attention includes: dynamically adjusting the channel weight through global average pooling GAP and fully connected layers to enhance important feature channels. The expression is:

[0058]

[0059] W c =σ(W2·δ(W1·F gap ))

[0060] Where, F gap is the feature map after global average pooling of the fully connected layer, W c is the feature channel enhanced by the fully connected layer after multi-scale cost processing, H is the number of horizontal categories of the feature map, W is the number of vertical categories of the feature map of the fully connected layer, F(i,j) is the total feature map under the number of horizontal categories and vertical categories of the fully connected layer, F is the input feature map of the fully connected layer, σ is the Sigmoid function, δ is the ReLU activation function, W1 and W2 are the parameters of the fully connected layer;

[0061] The spatial attention includes: using convolutional layers to generate a spatial weight map to highlight biological regions, expressed as:

[0062] W s =σ(f 3×3 [F avg ,F max ])

[0063] Where W s is the spatial weight value, f 3×3 is a 3×3 convolution, F avg ,F max They are channel average and maximum pooling results respectively;

[0064] The feature fusion includes combining channels with spatial attention weights to enhance features, which can be expressed as:

[0065] F enhanced =W c W s ·F

[0066] Where, F enhanced is the feature enhancement value after combining the channel and spatial attention weights, and F is the input feature map of the fully connected layer;

[0067] The optimization loss function adopts an improved composite loss function Composite Loss; including classification task loss and quantity estimation loss;

[0068] The classification task loss uses an improved Focal Loss to reduce the weight of easy-to-classify samples and alleviate category imbalance. The expression is:

[0069]

[0070] Where, L cls is the classification task loss value, p i is the predicted probability, y i is the true label, a i is the category weight;

[0071] The estimated losses include:

[0072] For dense biological areas, adaptive Huber Loss is used to balance outlier robustness and convergence speed, expressed as:

[0073]

[0074] Where, L count is the estimated loss value, n pred is the predicted value of dense biological outliers, n gt is the actual value of dense biological outliers, δ is the ReLU activation function, δ is dynamically adjusted according to training and decays with the number of iterations.

[0075] In step S2, data is transmitted to the communication module of the virtual anchor unmanned platform using an armored cable, and the underwater acoustic communication protocol includes the Modbus submarine communication protocol;

[0076] Perform error control on data transmitted via armored cables, detect and correct erroneous information that occurs during data transmission. By adding a cyclic redundancy check code (CRC) to the data, the receiver can detect whether errors occur during data transmission.

[0077] If an error is detected, the receiver requests the sender to resend the data;

[0078] When the receiver detects an error, it sends a request to the sender, asking the sender to resend the erroneous data packet. After receiving the request, the sender resends the packet until the receiver receives it correctly. When sending data, in addition to sending the original data, additional error correction codes are also sent. The receiver corrects the errors in the data based on these error correction codes.

[0079] In step S3, the information transmission control method within the underwater observation system includes:

[0080] Step 1: Determine the startup time. If yes, proceed to the next step. If no, re-determine.

[0081] Step 2: Start taking photos and start the acoustic Doppler current profiler ADCP;

[0082] Step 3: Perform 4G backhaul;

[0083] Step 4: If the transmission is successful, go to sleep or return to step 1; if the transmission fails, determine whether to resend the timeout;

[0084] If the resend timeout succeeds, the system goes into sleep mode or returns to step 1. If the resend timeout fails, the system returns to step 3 and performs 4G backhaul.

[0085] In step 4, the reissue timeout includes:

[0086] Step 4.1, timeout threshold setting;

[0087] Dynamic timeout calculation, adjusting the timeout based on historical network delay statistics:

[0088] T timeout =μRTT+3σRTT

[0089] Where, T timeout is the timer period, μRTT is the average round trip time, and σRTT is the standard deviation, ensuring coverage of 95% of the fluctuation range;

[0090] Step 4.2, reissue trigger conditions, including:

[0091] The first backhaul fails. When the 4G backhaul in step 3 does not receive an ACK, the retransmission process is triggered;

[0092] Retry queue management: data to be resent is stored in a persistent queue and marked as "pending retransmission";

[0093] Step 4.3, reissue execution process;

[0094] (1) Start the reissue timer and set the timer period to T timeout ; Start the asynchronous sending thread to try to resend data and listen for ACK response;

[0095] (2) Waiting for ACK or timeout, including: when ACK is successfully received, clearing the corresponding data in the queue, terminating the timer, entering sleep mode or returning to step 1; when ACK is not received after timeout, it is determined that the retransmission timeout has failed and the retry strategy is triggered;

[0096] Step 4.4, retry strategy optimization, including: using the exponential backoff algorithm to gradually increase the retry interval to avoid network congestion. The exponential backoff algorithm calculation formula is:

[0097] T retry =min(T base ×2 n-1 ,T max )

[0098] Where, T retry Optimize the value for the retry strategy, T base is the basic retry interval, T max is the maximum retry interval, T base =10s,T max =3600s, where n is the current number of retries.

[0099] Another object of the present invention is to provide an underwater online intelligent monitoring and analysis system, which implements the underwater online intelligent monitoring and analysis method, and the system includes:

[0100] The virtual anchored unmanned platform is used to stay on the sea surface, provide power to the seabed observation system and the underwater observation system, and transmit the information processed by the seabed observation system and the underwater observation system back to the land base station;

[0101] The seabed observation system transmits data to the virtual anchor unmanned platform via an armored cable, uses the onboard acoustic Doppler current profiler (ADCP) to measure seabed currents, obtain the speed and direction of water flow, and uses the onboard binocular camera to determine the size of underwater organisms;

[0102] The underwater observation system uses the image recognition algorithm within the underwater observation system to perform image pattern recognition and classification on underwater organisms, and identify the types and quantities of underwater organisms; and in the communication module, information is transmitted back to the land base station, and the information transmission control method is used to determine whether the returned information is successful.

[0103] The virtual anchor unmanned platform is composed of a buoy body, a communication module, a solar panel, a battery pack, a float, a counterweight and a supply-end cable;

[0104] The buoy body is responsible for carrying the equipment and keeping it on the sea surface;

[0105] The communication module is responsible for transmitting the collected data back to the land base station;

[0106] The energy supply module includes solar panels and battery packs. When solar power supply is insufficient, the battery pack provides backup power.

[0107] The float ball is installed as a whole with the buoy body to increase buoyancy;

[0108] Counterweights are used to increase stability;

[0109] Supply-side cable, connecting the buoy to the seabed observation system, transmitting data and power;

[0110] The seabed observation system includes: an acoustic Doppler current profiler ADCP, which is used for seabed water current observation;

[0111] Four-color lights for providing visual signals at night or in low visibility conditions;

[0112] Binocular camera, used to observe the seabed environment through stereo imaging technology;

[0113] Ultraviolet light, used for specific seabed biological observation or underwater navigation;

[0114] Batteries, used to provide electricity;

[0115] Collection board warehouse, used to store and process collected data;

[0116] The seabed base, which serves as a foundation, is fixed to the seabed;

[0117] Demand-side cables: connect the seabed observation system with the supply-side cables to transmit data and power;

[0118] The demand-side cable corresponds to the supply-side cable, running from the junction box to the seabed observation system. It is responsible for transmitting command data and power to the seabed observation system and receiving observation data.

[0119] The underwater observation system includes: an image recognition algorithm module, which is operated by a software system to analyze and identify underwater images;

[0120] The supporting algorithm software module for running the image recognition algorithm is responsible for processing and analyzing image data and identifying underwater targets.

[0121] Combining all of the above technical solutions, the present invention achieves the following beneficial effects: Using a virtual anchored unmanned platform as a signal and energy supply, the present invention utilizes a docking unit to transmit data via armored cables to a seafloor observation system for data transmission and energy supply. Furthermore, the seafloor observation system can be equipped with various sensors, particularly ADCP sensors for current measurement and binocular cameras for determining the size of underwater organisms. The underwater observation system utilizes machine vision algorithms for image pattern recognition and classification, and counts underwater organisms, ultimately achieving the calculation of the number and size of underwater organisms.

[0122] The present invention also solves the following problems:

[0123] Data transmission issues: Traditional underwater image recognition systems often experience data distortion in complex sea conditions, leading to signal interruptions and information loss. This problem is primarily due to the lack of a reliable data transmission mechanism and the instability of signal transmission in harsh marine environments. This invention utilizes a virtual anchored unmanned platform as the signal and energy supply, using armored cables for data transmission and energy supply, thereby enhancing the stability and reliability of data transmission.

[0124] Energy consumption and supply issues: Existing systems consume excessively high amounts of energy and provide intermittent power. This is often due to a lack of efficient energy management and supply systems, as well as the difficulty of maintaining a continuous power supply in underwater environments. This invention, by integrating solar panels and batteries, provides a sustainable energy solution to support the long-term operation of the buoy.

[0125] Comprehensive Observation Capabilities: Traditional underwater image recognition systems can only recognize images but cannot comprehensively observe the underwater environment. This limits the system's ability to comprehensively monitor changes in the marine environment. The underwater observation system of the present invention not only incorporates image recognition algorithms but also utilizes a seabed observation system equipped with multiple sensors, such as ADCP sensors for current measurement and binocular cameras for determining the size of underwater organisms, thereby enabling comprehensive monitoring and analysis of the underwater environment.

[0126] In summary, the present invention can achieve lancelet image recognition in complex sea conditions, improving operational efficiency and safety. Furthermore, the method has good adaptability and stability, making it suitable for a variety of offshore operational scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0127] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure;

[0128] Figure 1 Schematic diagram of an underwater online intelligent monitoring and analysis system provided by an embodiment of the present invention;

[0129] Figure 2 This is a schematic diagram of the stereo vision principle of a binocular camera provided by an embodiment of the present invention;

[0130] Figure 3 This is a schematic diagram of an image recognition algorithm provided by an embodiment of the present invention;

[0131] Figure 4 This is a schematic diagram of the information transmission control method within the underwater observation system provided by an embodiment of the present invention;

[0132] Figure 5 This is a diagram of the error control principle provided by an embodiment of the present invention;

[0133] In the picture: 1. Virtual anchored unmanned platform; 2. Seabed observation system; 3. Underwater observation system. DETAILED DESCRIPTION

[0134] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0135] Example 1, as Figure 1 As shown, the underwater online intelligent monitoring and analysis system provided by the embodiment of the present invention includes: a virtual anchored unmanned platform 1, a seabed observation system 2, and an underwater observation system 3.

[0136] The virtual anchor unmanned platform 1 is used to remain on the sea surface and provide power to the seabed observation system 2 and the underwater observation system 3, and is also used to transmit information processed by the seabed observation system 2 and the underwater observation system 3 back to the land base station;

[0137] The seabed observation system 2 transmits data with the virtual anchor unmanned platform 1 through an armored cable.

[0138] The onboard acoustic Doppler current profiler (ADCP) is used to measure seabed currents, obtain the speed and direction of water flow, and the onboard binocular camera is used to determine the size of underwater organisms.

[0139] Underwater observation system 3 uses the image recognition algorithm within the underwater observation system to perform image pattern recognition and classification on underwater organisms, and identify the types and quantities of underwater organisms; and in the communication module, when information is transmitted back to the land base station, the information transmission control method is used to determine whether the returned information is successful.

[0140] Exemplarily, the virtual anchor unmanned platform 1 is composed of a buoy body, a communication module, a solar panel, a battery pack, a float, a counterweight, and a supply-end cable;

[0141] Seabed observation system 2 consists of a small wide dragon, four-color lights, binocular cameras, ultraviolet lights, batteries, collection panels, seabed bases and demand-side cables;

[0142] The underwater observation system 3 is mainly composed of an image recognition algorithm module and a supporting algorithm software module for running the image recognition algorithm.

[0143] In the virtual anchored unmanned platform 1, the buoy body: as the main body of the virtual anchored unmanned platform, the buoy body is responsible for carrying other equipment and keeping it on the sea surface.

[0144] Communication module: responsible for transmitting the collected data back to the land base station.

[0145] Energy supply module (power supply system): includes solar panels and battery packs. When solar power supply is insufficient, the battery pack will provide backup power.

[0146] Float: Installed as a whole with the buoy body to increase buoyancy and ensure the stability of the platform on the sea surface.

[0147] Counterweight: Increase the stability of the virtual anchored unmanned platform to prevent it from being washed away by the current.

[0148] Supply-side cable (drag cable): connects the buoy body and the seabed observation system 2, and plays the role of transmitting data and power.

[0149] The 1.2-meter buoy serves as the energy supply module and data transmission platform. The seabed observation system 2 is powered by the energy supply module. The buoy is powered by a 3*100Ah battery pack.

[0150] The front end of the supply-end cable is connected to the energy supply module and the communication module through a slip ring, and the rear end is connected to the collection board warehouse of the seabed observation system 2 through a drag cable connector; an anchor chain is also installed on the rear end of the supply-end cable.

[0151] For example, in the seabed observation system 2, the acoustic Doppler current profiler ADCP (small wide dragon) refers to a small seabed observation device.

[0152] Four-color lights: used to provide visual signals at night or in low visibility conditions.

[0153] Binocular camera: Used to observe the seabed environment through stereo imaging technology.

[0154] Ultraviolet light: may be used for specific seabed biological observation or underwater navigation.

[0155] Batteries: Provide power for the seafloor observation system.

[0156] Collection board silo (data junction box): used to store and process collected data.

[0157] Seabed base (base platform): serves as the base of the seabed observation system and is fixed on the seabed.

[0158] Demand-side cables: connect the seabed observation system with the supply-side cables to transmit data and power.

[0159] For example, the system is equipped with one underwater binocular camera, one underwater light, and one ADCP for underwater monitoring. The light and binocular camera are powered by the buoy, while the ADCP is self-powered. A data connection box is installed on the seabed (base platform) for data aggregation and transmission.

[0160] In the data link, the link uses the supply-end cable (drag cable) as the power supply and communication cable; a slip ring is installed on the top to prevent the cable from being twisted and broken.

[0161] In terms of recovery and maintenance: additional lifting points are added to the seabed base (base platform). Considering the problem of siltation caused by long-term deployment, divers can be arranged to install underwater lifting points and lift the platform when necessary.

[0162] Regarding power consumption, the buoy is powered by a 3x100Ah battery pack. Currently, data (including image data and flow measurement data collected by the ADCP) is transmitted once a day. Without effective sunlight and charging, the estimated operating time is 360 days.

[0163] Exemplarily, in the underwater observation system 3, the image recognition algorithm module: runs through the software system to analyze and recognize underwater images.

[0164] The supporting algorithm software module for running the image recognition algorithm is responsible for processing and analyzing image data and identifying underwater targets.

[0165] In embodiment 2, the present invention provides an underwater online intelligent monitoring and analysis method, which includes:

[0166] S1: Use the acoustic Doppler current profiler (ADCP) onboard the seabed observation system to measure seabed currents and obtain the speed and direction of the water flow. Use the binocular camera onboard the seabed observation system to determine the size of underwater organisms. Use the image recognition algorithm within the underwater observation system to perform image pattern recognition and classification on underwater organisms to identify the species and number of underwater organisms.

[0167] S2, based on all acquired data and image information, uses the armored cable to transmit data to the communication module of the virtual anchor unmanned platform, and performs error control on the data transmitted by the armored cable to detect and correct errors in the data transmission process;

[0168] S3, when the communication module transmits information back to the land base station, the information transmission control method inside the underwater observation system is used to determine whether the returned information is successful. If successful, the information is returned; otherwise, steps S1-S2 are repeated until the information is successfully returned.

[0169] For example, in step S1, the acoustic Doppler current profiler ADCP measures the seabed flow by emitting sound waves and receiving reflected signals. When the sound waves propagate in the water, the Doppler frequency shift caused by the water flow is measured, and the speed and direction of the water flow are calculated.

[0170] Specifically, an ADCP (Acoustic Doppler Current Profiler) is a commonly used marine current measurement device. It measures current velocity by emitting sound waves and receiving the reflected signals. When sound waves propagate through water, they undergo a Doppler frequency shift due to the influence of the water flow. By measuring this frequency shift, the ADCP can calculate the speed and direction of the water flow. ADCPs typically have multiple sound wave transmitters that can simultaneously measure current velocities at different depths, thereby providing current velocity profile data. To improve measurement accuracy, ADCPs may employ signal processing algorithms such as beamforming and signal coherence processing.

[0171] For example, the present invention uses beamforming and signal coherence processing algorithms to measure flow velocities at different depths. Traditional methods use multi-beam radial velocity to calculate three-dimensional flow velocities. The present invention's improved beamforming and signal coherence processing algorithm combines signal processing optimization, error compensation, and coordinate system conversion correction to measure flow velocities at different depths. Specifically, the following methods are used:

[0172] S101, data acquisition and preprocessing;

[0173] Acoustic wave transmission and reception: ADCP transmits multiple fixed-frequency acoustic beams (usually 4 beams, Janus configuration) into the water body, receives the reflected signals and records the Doppler frequency shift.

[0174] Frequency shift extraction uses fast Fourier transform (FFT) or adaptive filtering algorithms to extract frequency shift and suppress noise (such as ocean environmental noise or instrument electronic noise).

[0175] Time-depth stratification: the echo signal is stratified by time window, combined with the sound velocity profile (taking into account the real-time changes in temperature, salinity and pressure), to collect the emission frequency and beam tilt angle of each depth unit;

[0176] S102, radial velocity calculation;

[0177] The radial velocity calculation formula is:

[0178]

[0179] Where, v r is the radial velocity, Δf is the Doppler shift, c is the speed of sound, f0 is the transmitting frequency, and θ is the beam tilt angle;

[0180] S103, three-dimensional flow velocity solution;

[0181] The traditional method projects the radial velocities of the four beams into the instrument coordinate system (East-North-Sky, ENU), while the present invention solves the velocity components u, V, and w through an improved error-weighted least squares method.

[0182] The improved error-weighted least squares method is formulated as follows: considering the differences in the signal-to-noise ratio (SNR) of each beam, introducing the weight matrix W, and optimizing the velocity component solution:

[0183]

[0184] Where A is the beam pointing matrix, V r is the radial velocity vector, * T is the transposed matrix;

[0185] S104, direction calculation and post-processing;

[0186] Flow velocity direction synthesis:

[0187] The horizontal velocity vector of the eastward velocity u and the northward velocity v is the direction angle for:

[0188]

[0189] Where, It is the geomagnetic declination compensation value (needs to be updated in real time in combination with the geomagnetic field model).

[0190] In step S1, determining the size of underwater organisms using a binocular camera carried by a seabed observation system includes:

[0191] (1) Calibrate the binocular camera to determine the internal and external parameters of the camera;

[0192] (2) By acquiring the left and right images, feature extraction, feature matching, and disparity calculation are performed to achieve the perception of the depth information of the object;

[0193] (3) Reconstruct the three-dimensional scene based on the depth information.

[0194] Specifically, the role of the binocular camera is to obtain the three-dimensional geometric information of the object based on the principle of stereo vision. The principle of binocular vision is based on the concept of parallax, that is, the pixel displacement caused by the difference in viewpoint when the left and right cameras observe the same object. By calculating the parallax, the depth information of the object can be inferred. In binocular vision, the left camera is usually called the main camera and the right camera is called the auxiliary camera. First, the binocular camera needs to be calibrated to determine the internal and external parameters of the camera. Then, by obtaining the left and right images, feature extraction, feature matching and parallax calculation steps are performed to realize the perception of the depth information of the object. Finally, the three-dimensional scene is reconstructed based on the depth information. Among them, the principle diagram of stereo vision of the binocular camera is as follows Figure 2 shown.

[0195] Exemplarily, in step (2), the perception of the depth information of the object is realized by using an improved stereo matching algorithm, specifically including:

[0196] Step 1, feature extraction (improved robust descriptor);

[0197] A deep learning-based feature extraction network (such as SuperPoint) outputs feature point coordinates and descriptors:

[0198]

[0199] Where F(I) is the feature extraction image, x i is the horizontal coordinate of the feature point, y i is the ordinate of the feature point, is the feature point pixel, d i For high-dimensional descriptors, cross-view consistency is optimized through contrastive learning;

[0200] Step 2: Use graph neural network-based matching optimization to perform feature matching (improved graph optimization matching);

[0201] The matching optimization based on graph neural network includes:

[0202] Construct a graph structure of feature points of the left and right images, where nodes are feature points and edges are descriptor similarities;

[0203] Aggregate neighborhood information through the Graph Attention Network (GAT) to optimize the matching score:

[0204]

[0205] Where s ij To optimize the matching score, GAT() is an aggregation neighborhood function. is the i-th high-dimensional descriptor of the left neighborhood, is the jth high-dimensional descriptor of the right neighborhood;

[0206] Differentiable soft matching (SoftMatch) is used to generate a probabilistic matching matrix, and epipolar constraints are combined to screen out inliers.

[0207] Step 3: Disparity calculation using improved cost aggregation and optimization;

[0208] Improved cost aggregation and optimization include:

[0209] Multi-scale cost volume construction, generating multi-scale feature maps through CNN and constructing 3D cost volume C(d,x,y);

[0210] A dynamic path-weighted SGM is adopted, which includes the introduction of edge-aware weights in the smoothing term of the traditional SGM:

[0211]

[0212] Where E(d) is the smoothing term of SGM, λ is the edge perception weight, p and q are the p-th and q-th paths respectively, N is the number of paths, and w pq is the edge-aware eigenvalue of the pth and qth paths, d p is the high-dimensional descriptor of the p-th path, d q is the high-dimensional descriptor of the qth path; w pq =exp(-γ||I p -I q || 2 ) is adaptively adjusted according to the image gradient, γ is the focusing factor, exp() is the image gradient adaptive adjustment function, I p is the feature point of the p-th path, I q is the characteristic point of the qth path;

[0213] Sub-pixel precision optimization, using parabola interpolation or Newton iteration to optimize parallax:

[0214]

[0215] Where, d subpixel is the sub-pixel accuracy optimization value, C is the multi-scale cost, and d is a high-dimensional descriptor;

[0216] Exemplarily, in step S1, performing image pattern recognition and classification of underwater organisms using an image recognition algorithm within the underwater observation system includes:

[0217] First, the captured image is preprocessed by denoising and contrast enhancement;

[0218] Then the image features are extracted through the convolutional neural network CNN;

[0219] Next, the classifier is used to introduce an attention mechanism and optimize the loss function to analyze the features and identify the types and numbers of underwater organisms.

[0220] Specifically, underwater organism recognition algorithms are usually based on machine vision technology, and use image recognition algorithms to perform pattern recognition and classification of underwater organisms. These algorithms may use deep learning models, such as the YOLO (You Only LookOnce) series, which can achieve real-time object detection and recognition. In order to adapt to the particularities of the underwater environment, such as light changes and scattering of water bodies, image recognition algorithms introduce attention mechanisms and optimize loss functions. During the recognition process, the captured image is first preprocessed, such as denoising and contrast enhancement, and then the image features are extracted through a convolutional neural network (CNN). Next, a classifier is used to analyze the features to identify the types and quantities of underwater organisms. For size measurement, image recognition may use stereo images obtained by binocular cameras to calculate the three-dimensional size of the organism through stereo vision technology. The principle diagram of the image recognition algorithm is as follows Figure 3 shown.

[0221] Exemplarily, the classifier is used to introduce an attention mechanism and optimize a loss function to analyze features and identify the types and quantities of underwater organisms. The attention mechanism adopts an improved hybrid attention module (HAM), including channel attention, spatial attention, and feature fusion.

[0222] The channel attention includes:

[0223] Dynamically adjust channel weights through global average pooling (GAP) and fully connected layers to enhance important feature channels:

[0224]

[0225] W c =σ(W2·δ(W1·F gap ))

[0226] Where, F gap is the feature map after global average pooling of the fully connected layer, W c is the feature channel enhanced by the fully connected layer after multi-scale cost processing, H is the number of horizontal categories of the feature map, W is the number of vertical categories of the feature map of the fully connected layer, F(i,j) is the total feature map under the number of horizontal categories and vertical categories of the fully connected layer, F is the input feature map of the fully connected layer, σ is the Sigmoid function, δ is the ReLU activation function, W1 and W2 are the parameters of the fully connected layer;

[0227] The spatial attention includes:

[0228] The convolutional layer is used to generate a spatial weight map to highlight the biological region. The expression is:

[0229] W s =σ(f 3×3 [F avg ,F max ])

[0230] Where W s is the spatial weight value, f 3×3 is a 3×3 convolution, F avg ,F max They are channel average and maximum pooling results respectively;

[0231] The feature fusion includes combining channels with spatial attention weights to enhance features, which can be expressed as:

[0232] F enhanced =W c W s ·F

[0233] Where, F enhanced is the feature enhancement value after combining the channel and spatial attention weights, and F is the input feature map of the fully connected layer;

[0234] The optimization loss function adopts an improved composite loss function (Composite Loss); including classification task loss and quantity estimation loss;

[0235] The classification task loss uses an improved Focal Loss to reduce the weight of easy-to-classify samples and alleviate category imbalance. The expression is:

[0236]

[0237] Where, L cls is the classification task loss value, p i is the predicted probability, y i is the true label, a i is the category weight (inversely proportional to the number of samples);

[0238] The estimated losses include:

[0239] For dense biological areas, adaptive Huber Loss is used to balance outlier robustness and convergence speed, expressed as:

[0240]

[0241] Where, L count is the estimated loss value, n pred is the predicted value of dense biological outliers, n gtis the actual value of dense biological outliers, δ is the ReLU activation function, δ is dynamically adjusted according to training and decays with the number of iterations.

[0242] Exemplarily, in the communication module for transmitting data to the virtual anchored unmanned platform using the armored cable in step S2, the underwater acoustic communication protocol includes the Modbus submarine communication protocol;

[0243] Error control is performed on the data transmitted via the armored cable to detect and correct the error information that occurs during the data transmission process, including: by adding a cyclic redundancy check code CRC to the data, the receiver detects whether there is an error in the data transmission process;

[0244] If an error is detected, the receiver requests the sender to resend the data;

[0245] When the receiver detects an error, it sends a request to the sender, asking the sender to resend the erroneous data packet. After receiving the request, the sender resends the packet until the receiver receives it correctly. When sending data, in addition to sending the original data, additional error correction codes are also sent. The receiver corrects the errors in the data based on these error correction codes.

[0246] Specifically, in terms of data transmission technology, the virtual anchor unmanned platform 1 transmits data to the seabed observation system 2 via an armored cable. The armored cable is a sturdy cable commonly used in marine environments that can withstand seawater corrosion and seabed pressure.

[0247] Because electromagnetic waves propagate much less efficiently underwater than sound waves, underwater acoustic communication has become the primary method of underwater communication. Underwater acoustic communication protocols typically include modulation and demodulation techniques, signal processing algorithms, and data link layer protocols to ensure reliable data transmission in complex underwater environments.

[0248] Data transmission uses multiple communication protocols, such as the Modbus dedicated submarine communication protocol, to ensure stable and reliable data transmission. At the same time, error control and flow control technologies may be used to improve transmission efficiency and reduce data loss.

[0249] Error control is a mechanism used to detect and correct erroneous information that occurs during data transmission. Its purpose is to ensure that the data received by the receiver is exactly the same as the data sent by the sender, and that errors can be corrected even if they occur during transmission. By adding a cyclic redundancy check code (CRC) to the data, the receiver can detect whether errors occur during data transmission. If an error is detected, the receiver can request the sender to resend the data. When the receiver detects an error, it will send a request to the sender, requesting the sender to resend the erroneous data packet. After receiving the request, the sender will resend the packet until the receiver receives it correctly. When sending data, in addition to sending the original data, some additional error correction codes are also sent. The receiver can correct errors in the data based on these error correction codes, such as forward error correction (FEC). For example Figure 5 Error control principle diagram.

[0250] For example, in step S3, Figure 4 As shown, the information transmission control method within the underwater observation system includes:

[0251] Step 1: Determine the startup time. If yes, proceed to the next step. If no, re-determine.

[0252] Step 2: Start taking photos and start the acoustic Doppler current profiler ADCP;

[0253] Step 3: Perform 4G backhaul;

[0254] Step 4: If the backhaul is successful, go to sleep or return to step 1; if the backhaul fails, determine whether to resend the timeout. If the resend timeout is successful, go to sleep or return to step 1. If the resend timeout fails, return to step 3 and re-4G backhaul.

[0255] Exemplarily, the reissue timeout includes:

[0256] Step 4.1, timeout threshold setting;

[0257] Dynamic timeout calculation, adjusting the timeout based on historical network delay statistics (such as sliding window average RTT):

[0258] T timeout =μRTT+3σRTT

[0259] Where, T timeout is the timer period, μRTT is the average round trip time, and σRTT is the standard deviation, ensuring coverage of 95% of the fluctuation range;

[0260] Step 4.2, reissue trigger conditions, including:

[0261] The first return fails. When the 4G return in step 3 does not receive an ACK (or the HTTP status code is not 200), the resend process is triggered;

[0262] Retry queue management stores the data to be resent in a persistent queue (such as Flash storage) and marks it as "pending retransmission".

[0263] Step 4.3, reissue execution process;

[0264] 1) Start the reissue timer and set the timer period to T timeout ;

[0265] Start the asynchronous sending thread to try to resend data and listen for ACK responses at the same time;

[0266] 2) Waiting for ACK or timeout, including:

[0267] When ACK is successfully received, the corresponding data in the queue is cleared, the timer is terminated, and the system goes into sleep mode or returns to step 1.

[0268] If no ACK is received after the timeout, it is determined that the retransmission timeout has failed and the retry strategy is triggered;

[0269] Step 4.4, retry strategy optimization, includes:

[0270] Use the exponential backoff algorithm to gradually increase the retry interval to avoid network congestion. The calculation formula of the exponential backoff algorithm is:

[0271] T retry =min(T base ×2 n-1 ,T max )

[0272] Where, T retry Optimize the value for the retry strategy, T base is the basic retry interval, T max is the maximum retry interval, T base =10s,T max =3600s, where n is the current number of retries.

[0273] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0274] The above description is only a preferred specific implementation method of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. An underwater online intelligent monitoring and analysis method, characterized in that: The method comprises the following steps: S1: Use the acoustic Doppler current profiler (ADCP) onboard the seabed observation system to measure seabed currents and obtain the speed and direction of the water flow. Use the binocular camera onboard the seabed observation system to determine the size of underwater organisms. Use the image recognition algorithm within the underwater observation system to perform image pattern recognition and classification on underwater organisms to identify the species and number of underwater organisms. S2, based on all acquired data and image information, uses the armored cable to transmit data to the communication module of the virtual anchor unmanned platform, and performs error control on the data transmitted by the armored cable to detect and correct errors in the data transmission process; S3, when the communication module transmits information back to the land base station, the information transmission control method inside the underwater observation system is used to determine whether the returned information is successful. If successful, the information is returned; otherwise, steps S1-S2 are repeated until the information is successfully returned.

2. The underwater online intelligent monitoring and analysis method according to claim 1, characterized in that: In step S1, the acoustic Doppler current profiler (ADCP) measures seabed currents by emitting sound waves and receiving reflected signals. When the sound waves propagate through the water, the Doppler frequency shift caused by the water flow is measured, and the speed and direction of the water flow are calculated. The acoustic Doppler current profiler (ADCP) has multiple sound wave transmitters and uses beamforming and signal coherence processing algorithms to measure flow velocities at different depths and provide flow profile data.

3. The underwater online intelligent monitoring and analysis method according to claim 2, characterized in that: The beamforming and signal coherence processing algorithms include: S101, data acquisition and preprocessing; Acoustic wave transmission and reception: ADCP transmits multiple fixed-frequency acoustic beams into the water, receives the reflected signals and records the Doppler shift; Frequency shift extraction, using fast Fourier transform FFT or adaptive filtering algorithm to extract frequency shift; Time-depth stratification: stratify the echo signal by time window, combine it with the sound velocity profile, and collect the emission frequency and beam tilt angle of each depth unit; S102, radial velocity calculation; The radial velocity calculation formula is: Where, v r is the radial velocity, Δf is the Doppler shift, c is the speed of sound, f0 is the transmitting frequency, and θ is the beam tilt angle; S103, three-dimensional flow velocity solution; The velocity components u, v, w are solved by the improved error-weighted least squares method; The improved error-weighted least squares method takes into account the difference in signal-to-noise ratio (SNR) of each beam, introduces the weight matrix W, and optimizes the solution of the velocity component. The expression is: Where A is the beam pointing matrix, V r is the radial velocity vector, * T is the transposed matrix; S104, direction calculation and post-processing; Flow velocity direction synthesis: The horizontal velocity vector of the eastward velocity u and the northward velocity v is the direction angle for: Where, is the geomagnetic declination compensation value.

4. The underwater online intelligent monitoring and analysis method according to claim 1, characterized in that: In step S1, the size of underwater organisms is determined using a binocular camera carried by a seabed observation system, including: (1) Calibrate the binocular camera to determine the internal and external parameters of the camera; (2) By acquiring the left and right images, performing feature extraction, feature matching, and disparity calculation, the depth information of the object can be perceived. The improved stereo matching algorithm is used to perceive the depth information of the object, specifically including: Step 1: Feature extraction: Based on the deep learning feature extraction network, the feature point coordinates and descriptors are output. The expression is: Where F(I) is the feature extraction image, x i is the horizontal coordinate of the feature point, y i is the ordinate of the feature point, is the feature point pixel, d i For high-dimensional descriptors, cross-view consistency is optimized through contrastive learning; Step 2: Use graph neural network-based matching optimization to perform feature matching; Matching optimization based on graph neural networks includes: building a graph structure of feature points of the left and right images, with nodes being feature points and edges being descriptor similarities; The graph attention network GAT aggregates neighborhood information and optimizes the matching score, which is expressed as: Where s ij To optimize the matching score, GAT() is an aggregation neighborhood function. is the i-th high-dimensional descriptor of the left neighborhood, is the jth high-dimensional descriptor of the right neighborhood; Use differentiable soft matching SoftMatch to generate a probability matching matrix and combine it with polar constraints to filter out inliers; Step 3: Disparity calculation using improved cost aggregation and optimization; Improved cost aggregation and optimization include: Multi-scale cost volume construction, generating multi-scale feature maps through CNN and constructing 3D cost volume C(d,x,y); Dynamic path weighted SGM is adopted, and edge-aware weight is introduced into the smoothing term of SGM, which is expressed as: Where E(d) is the smoothing term of SGM, λ is the edge perception weight, p and q are the p-th and q-th paths respectively, N is the number of paths, and w pq is the edge-aware eigenvalue of the pth and qth paths, d p is the high-dimensional descriptor of the p-th path, d q is the high-dimensional descriptor of the qth path; w pq =exp(-γ||I p -I q ‖ 2 ) is adaptively adjusted according to the image gradient, γ is the focusing factor, exp() is the image gradient adaptive adjustment function, I p is the feature point of the p-th path, I q is the characteristic point of the qth path; Sub-pixel precision optimization, using parabola interpolation or Newton iteration to optimize parallax: Where, d subpixel is the sub-pixel accuracy optimization value, C is the multi-scale cost, and d is a high-dimensional descriptor; (3) Reconstruct the three-dimensional scene based on the depth information.

5. The underwater online intelligent monitoring and analysis method according to claim 4, characterized in that: In step S1, the image recognition algorithm within the underwater observation system is used to perform image pattern recognition and classification of underwater organisms, including: First, the captured images are preprocessed with denoising and contrast enhancement; Then, the image features are extracted through the convolutional neural network CNN; Next, the classifier introduces an attention mechanism and optimizes the loss function to analyze the features and identify the types and numbers of underwater creatures. The attention mechanism uses an improved hybrid attention module (HAM), which includes channel attention, spatial attention, and feature fusion. The channel attention includes: dynamically adjusting the channel weight through global average pooling GAP and fully connected layers to enhance important feature channels. The expression is: W c =σ(W2·δ(W1·F gap )) Where, F gap is the feature map after global average pooling of the fully connected layer, W c is the feature channel enhanced by the fully connected layer after multi-scale cost processing, H is the number of horizontal categories of the feature map, W is the number of vertical categories of the feature map of the fully connected layer, F(i,j) is the total feature map under the number of horizontal categories and vertical categories of the fully connected layer, F is the input feature map of the fully connected layer, σ is the Sigmoid function, δ is the ReLU activation function, W1 and W2 are the parameters of the fully connected layer; The spatial attention includes: using convolutional layers to generate a spatial weight map to highlight biological regions, expressed as: W s =σ(f 3×3 [F avg ,F max ]) Where W s is the spatial weight value, f 3×3 is a 3×3 convolution, F avg ,F max They are channel average and maximum pooling results respectively; The feature fusion includes combining channels with spatial attention weights to enhance features, which can be expressed as: F enhanced =W c ·W s ·F Where, F enhanced is the feature enhancement value after combining the channel and spatial attention weights, and F is the input feature map of the fully connected layer; The optimization loss function adopts an improved composite loss function Composite Loss; including classification task loss and quantity estimation loss; The classification task loss uses an improved Focal Loss to reduce the weight of easy-to-classify samples and alleviate category imbalance. The expression is: Where, L cls is the classification task loss value, p i is the predicted probability, y i is the true label, a i is the category weight; The estimated losses include: For dense biological areas, adaptive Huber Loss is used to balance outlier robustness and convergence speed, expressed as: Where, L count is the estimated loss value, n pred is the predicted value of dense biological outliers, n gt is the actual value of dense biological outliers, δ is the ReLU activation function, δ is dynamically adjusted according to training and decays with the number of iterations.

6. The underwater online intelligent monitoring and analysis method according to claim 1, characterized in that: In step S2, data is transmitted to the communication module of the virtual anchor unmanned platform using an armored cable, and the underwater acoustic communication protocol includes the Modbus submarine communication protocol; Perform error control on data transmitted via armored cables, detect and correct erroneous information that occurs during data transmission. By adding a cyclic redundancy check code (CRC) to the data, the receiver can detect whether errors occur during data transmission. If an error is detected, the receiver requests the sender to resend the data; When the receiver detects an error, it sends a request to the sender, asking the sender to resend the erroneous data packet. After receiving the request, the sender resends the packet until the receiver receives it correctly. When sending data, in addition to sending the original data, additional error correction codes are also sent. The receiver corrects the errors in the data based on these error correction codes.

7. The underwater online intelligent monitoring and analysis method according to claim 1, characterized in that: In step S3, the information transmission control method within the underwater observation system includes: Step 1: Determine the startup time. If yes, proceed to the next step. If no, re-determine. Step 2: Start taking photos and start the acoustic Doppler current profiler ADCP; Step 3: Perform 4G backhaul; Step 4: If the transmission is successful, go to sleep or return to step 1; if the transmission fails, determine whether to resend the timeout; If the resend timeout succeeds, the system goes into sleep mode or returns to step 1. If the resend timeout fails, the system returns to step 3 and performs 4G backhaul.

8. The underwater online intelligent monitoring and analysis method according to claim 7, characterized in that: In step 4, the reissue timeout includes: Step 4.1, timeout threshold setting; Dynamic timeout calculation, adjusting the timeout based on historical network delay statistics: T timeout =μRTT+3σRTT Where, T timeout is the timer period, μRTT is the average round trip time, and σRTT is the standard deviation, ensuring coverage of 95% of the fluctuation range; Step 4.2, reissue trigger conditions, including: The first backhaul fails. When the 4G backhaul in step 3 does not receive an ACK, the retransmission process is triggered; Retry queue management: data to be resent is stored in a persistent queue and marked as "pending retransmission"; Step 4.3, reissue execution process; (1) Start the reissue timer and set the timer period to T timeout ; Start the asynchronous sending thread to try to resend data and listen for ACK response; (2) Waiting for ACK or timeout, including: when ACK is successfully received, clearing the corresponding data in the queue, terminating the timer, entering sleep mode or returning to step 1; when ACK is not received after timeout, it is determined that the retransmission timeout has failed and the retry strategy is triggered; Step 4.4, retry strategy optimization, including: using the exponential backoff algorithm to gradually increase the retry interval to avoid network congestion. The exponential backoff algorithm calculation formula is: T retry =min(T base ×2 n-1 ,T max ) Where, T retry Optimize the value for the retry strategy, T base is the basic retry interval, T max is the maximum retry interval, T base =10s,T max =3600s, where n is the current number of retries.

9. An underwater online intelligent monitoring and analysis system, characterized in that: The system implements the underwater online intelligent monitoring and analysis method according to any one of claims 1 to 8, and the system includes: The virtual anchor unmanned platform (1) is used to remain on the sea surface, provide power to the seabed observation system (2) and the underwater observation system (3), and transmit information processed by the seabed observation system (2) and the underwater observation system (3) back to the land base station; The seabed observation system (2) transmits data with the virtual anchor unmanned platform (1) via an armored cable, uses an onboard acoustic Doppler current profiler ADCP to measure seabed currents, obtains the speed and direction of the water flow, and uses an onboard binocular camera to determine the size of underwater organisms; The underwater observation system (3) uses the image recognition algorithm within the underwater observation system to perform image pattern recognition and classification on underwater organisms, and identifies the types and quantities of underwater organisms; and in the communication module, when information is transmitted back to the land base station, the information transmission control method is used to determine whether the returned information is successful.

10. The underwater online intelligent monitoring and analysis system according to claim 9, characterized in that: The virtual anchor unmanned platform (1) is composed of a buoy body, a communication module, a solar cell panel, a battery pack, a float, a counterweight block and a supply end cable; The buoy body is responsible for carrying the equipment and keeping it on the sea surface; The communication module is responsible for transmitting the collected data back to the land base station; The energy supply module includes solar panels and battery packs. When solar power supply is insufficient, the battery pack provides backup power. The float ball is installed as a whole with the buoy body to increase buoyancy; Counterweights are used to increase stability; Supply-side cable, connecting the buoy body with the seabed observation system (2), transmitting data and power; The seabed observation system (2) comprises: an acoustic Doppler current profiler ADCP, used for seabed water flow observation; Four-color lights for providing visual signals at night or in low visibility conditions; Binocular camera, used to observe the seabed environment through stereo imaging technology; Ultraviolet light, used for specific seabed biological observation or underwater navigation; Batteries, used to provide electricity; Collection board warehouse, used to store and process collected data; The seabed base, which serves as a foundation, is fixed to the seabed; Demand-side cables: connect the seabed observation system with the supply-side cables to transmit data and power; The demand-side cable corresponds to the supply-side cable, running from the junction box to the seabed observation system. It is responsible for transmitting command data and power to the seabed observation system and receiving observation data. The underwater observation system (3) includes: an image recognition algorithm module, which is operated by a software system to analyze and recognize underwater images; The supporting algorithm software module for running the image recognition algorithm is responsible for processing and analyzing image data and identifying underwater targets.

Images