A method and system for locating and targeting damaged netting in deep-sea cages

Abstract

This invention discloses a method and system for locating and targeting damaged netting in deep-sea cages, relating to the field of marine engineering technology. It simultaneously collects time-history data of netting tension and wave height; derives amplitude-frequency response operators for wave excitation and tension response, combines coherence function verification, and extracts the transfer function distortion rate by comparing with a healthy baseline state to trigger a local early warning; extracts the out-of-limit weights of abnormal measurement points, calculates the three-dimensional target coordinates of suspected damaged meshes, and sends them to an underwater robot; the underwater robot navigates to these target coordinates, uses a graph-based visual feature recognition algorithm to extract netting knots and anchor points to construct a topology map, and diagnoses netting breakage through topological anomalies. This invention employs the aforementioned method and system for locating and targeting damaged netting in deep-sea cages, utilizing frequency domain features to reveal structural stiffness degradation, effectively filtering out random sea state interference, and achieving a closed-loop electromechanical collaborative targeted inspection with high interference resistance, low computational consumption, and high efficiency.

Description

Technical Field

[0001] This invention relates to the field of marine engineering technology, and in particular to a method and system for locating and targeting damaged netting in deep-sea cages. Background Technology

[0002] Deep-sea cage aquaculture is a crucial measure for the structural transformation of aquaculture from nearshore to deep-sea farming, contributing to the sustainable development of marine fisheries, alleviating nearshore environmental pollution, and improving the quality of marine products. However, under the continuous influence of complex ocean currents and waves, the underwater netting systems of aquaculture cages are highly susceptible to damage. If not detected in time, mass escapes of farmed fish can occur, causing incalculable economic losses and potentially leading to invasive species and other ecological security issues. Therefore, developing more effective methods for preventing and controlling cage netting damage is a key technology for ensuring marine economic security.

[0003] Current methods for preventing and warning of damage to net cages and mesh mainly rely on the following technical approaches: First, using underwater robots (ROVs) to conduct large-scale underwater blind scanning inspections using a global parallel plowing or spiral path. This method suffers from low scanning efficiency, high power consumption, and a high risk of missed detections. Even when combined with advanced image processing algorithms to extract the mesh edges for defect classification, this method is highly dependent on the selection of ROV lenses and lacks a macroscopic guidance mechanism. Second, establishing hydrodynamic simulation models of net cages and mesh to predict vulnerable areas or comparing measured boundary force data with a large database of virtual mathematical models of mesh. However, under the influence of real and complex random ocean waves, the absolute values ​​of net cage and mesh fluctuate drastically in the time domain. Direct comparison based on time domain data is easily affected by sudden environmental changes, resulting in false alarms. At the same time, it requires extremely high computing power, making it difficult to efficiently deploy edge computing nodes at sea and truly achieve effective decoupling between environmental excitation and structural response.

[0004] Therefore, there is an urgent need for a method and system for locating and inspecting damaged netting in cages that is effective against environmental interference, consumes low computing resources, and can guide underwater robots to achieve precise targeted inspection. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for locating and targeting damaged netting in deep-sea cages, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a method for locating and targeting damaged netting in deep-sea cages, comprising the following steps: S1. Tension sensors are installed on the netting of the cage to collect tension time history data synchronously, and wave measurement equipment is installed around the cage to acquire wave height time history data synchronously. S2. Set the sliding time data window, calculate the sea state characteristic parameters based on the wave height time history data and calculate their variance. When the variance meets the preset environmental fluctuation discrimination conditions, it is determined that the sea state is in a stable period and the frequency domain decoupling calculation is triggered. Otherwise, it is determined to be a sea state change period and the alarm is filtered out. S3. During the period of stable sea conditions, the amplitude-frequency response operator between wave excitation and net tension response is calculated using spectral analysis theory, and a coherence function is introduced to test its effectiveness. If the coherence function is greater than the set threshold, the current amplitude-frequency response operator is determined to be effective. S4. Extract the main peak frequency and curve energy integral of the effective amplitude-frequency response operator in the main energy frequency band of the wave, and compare it with the pre-stored health benchmark transfer function. Calculate the transfer function distortion rate. If the distortion rate is greater than the safety diagnosis threshold, trigger a local early warning. S5. Extract the absolute physical coordinates and over-limit weights of the abnormal measuring points that trigger local early warning, and calculate the three-dimensional target relative coordinates of the suspected damaged mesh. S6. Send the three-dimensional target relative coordinates to the underwater robot. The underwater robot autonomously navigates to the three-dimensional target relative coordinates and starts the computer vision module to extract the set of knots and feet of the netting to construct an undirected topology graph. By judging the abnormality of the structural features of the topology graph, the physical breakage of the netting is diagnosed, and the damage assessment diagnosis image and coordinate data are transmitted back.

[0007] Preferably, S1 includes: S11. Set up several measuring points on the vertical steel ropes of the cage mesh, and arrange several tension sensors at the measuring points to form a set of measuring points. ; S12. At each measuring point, use a preset sampling rate. Synchronous data acquisition was performed to obtain the tensile time history data of the netting at each measuring point. ; S13. Deploy wave measurement equipment in the sea area surrounding the net cages to simultaneously acquire wave height time history data at corresponding times. ; S14. Collect the tensile time history data. With wave height time history data The original signal is filtered by a bandpass filter to remove high-frequency electrical noise and low-frequency long-period tidal drift, and an effective dynamic time sequence with zero mean is obtained.

[0008] Preferably, the process of determining that the sea state is in a stable period in S2 is as follows: S21, in any... Within a sliding time data window, the wave height time history signal is subjected to a discrete Fourier transform to obtain the wave's one-sided power spectral density, and the zero-order moment is calculated. With second moment ; S22. Calculate the effective wave height within this window. With average zero-crossing cycle ; S23, Statistical Continuity The variance of the effective wave height and the average zero-crossing period within each sliding time window and If and only if and At that time, it was determined that the sea conditions were in a period of stability, among which, This is a preset environmental extreme value threshold.

[0009] Preferably, the process of calculating the amplitude-frequency response operator and performing message verification in S3 is as follows: S31. Apply a smoothing Hanning window function to the zero-mean-normalized wave height discrete signal sequence and the tension discrete signal sequence. S32. Calculate the averaged wave power spectral density using the Welch average periodogram method. Tensile power spectral density and cross-spectral density ; S33. Solve the complex transfer function by the ratio of the cross spectral density to the wave self-power spectral density, and extract its magnitude as the amplitude-frequency response operator. ; S34. Calculate the coherence function If within the main energy frequency band of the wave If so, the current amplitude-frequency response operator is considered valid, where, This is the coherence threshold.

[0010] Preferably, the coherence function in S34 As shown in the following formula: .

[0011] Preferably, in step S4, the transfer function distortion rate is calculated using the following formula. : ; in, and These are the peak frequency and amplitude of the current effective transfer function, respectively; and These represent the main peak frequency and amplitude of the reference transfer function, respectively. These are the normalized weighting coefficients.

[0012] Preferably, in step S5, the three-dimensional target relative coordinates of the suspected damaged mesh are calculated using the following formula. : ; ; ; in, The absolute physical coordinates of the abnormal measurement points. The excessive weight of this abnormal measurement point. This is the spatial decay control factor.

[0013] Preferably, the graph-based local topological feature recognition algorithm in step S6 extracts the set of nodes in the mesh, including: The original video frames are acquired and converted to the LAB color space, the luminance channel is preserved, the local contrast is enhanced by the contrast-limited adaptive histogram equalization algorithm, and the high-frequency noise is smoothed by Gaussian bilateral filtering to obtain the enhanced image. The Harris corner detection operator is used to traverse and enhance the image to calculate feature values, and pixels with feature values ​​greater than a set threshold are extracted as candidate nodule pixel sets. A density-based spatial clustering algorithm is used to cluster the candidate nodule pixel set. The geometric center of each cluster is calculated as the effective netted nodule, forming a node set. .

[0014] Preferably, the graph theory-based local topological feature recognition algorithm in step S6 constructs an undirected topological graph from the set of footprints, including: Adaptive local binarization is applied to the enhanced image to separate the foreground mask, and a morphological thinning algorithm is used to refine the mesh foreground into a skeleton topology map with a single pixel width. With node set Starting from any node in the graph, perform an eight-neighbor search along a connected single-pixel path on the skeleton topology graph. If the tracing path successfully reaches another node and its length is within the tolerance range, then it is determined that there is a physical destination between the two points, forming an edge set. ; Based on node set Sum of edges Constructing an undirected topological graph of the local mesh When an abnormal decrease in the degree of multiple consecutive nodes is detected in an undirected topology graph, or when the number of edges constituting a closed mesh is found to have increased to the threshold of an irregular polygon through closed loop detection, it is determined that a network break has occurred in the undirected topology graph.

[0015] A system for locating and targeting damage to deep-sea cage mesh includes: The data synchronization acquisition module is used to synchronously acquire the tension time history data and the wave height time history data of the netting at each measuring point, and perform filtering preprocessing. The edge computing and frequency domain decoupling module is used to perform sea state stability verification and calculate the amplitude-frequency response operator, coherence function and transfer function distortion rate during the sea state stability period, thereby triggering local early warning; The spatial positioning and command generation module is used to extract the absolute physical coordinates and their over-limit weights of abnormal measurement points, calculate the three-dimensional target relative coordinates of suspected damaged mesh, and generate waypoint commands for the underwater robot. The underwater robot execution module is used to autonomously navigate to the target coordinates after receiving instructions. It uses a computer vision module to extract the knot set of the netting and construct an undirected topology graph with the target feet. It diagnoses physical breaks in the netting through topological anomalies and transmits damage assessment data back.

[0016] Therefore, the present invention employs the above-mentioned method and system for locating and targeting damage to deep-sea cage mesh, which has the following beneficial effects: (1) This method abandons the traditional approach of forcibly comparing measured time-domain data with a huge virtual database. It uses the frequency domain transfer function to directly reveal the local stiffness changes of the structure. No matter how violent the external random waves or ocean currents fluctuate, the inherent transfer function of the structure remains relatively stable when it is not damaged, thereby greatly filtering out false alarms caused by sea state fluctuations.

[0017] (2) The edge computing resources used in this method are low. Only one-dimensional spectral analysis and transfer function calculation are required within the sliding time window. There is no need to perform complex three-dimensional nonlinear hydrodynamic time-domain finite element solution. The algorithm is lightweight and can be easily deployed in the edge computing nodes of marine buoys or control platforms.

[0018] (3) This method also has highly efficient targeted guidance inspection, which completely changes the inefficient mode of underwater robots relying on preset parallel or spiral paths for global blind scanning. It uses a tension sensor array as a macro radar to directly calculate the three-dimensional coordinates of suspected damage and guide the underwater robot to the breach. The inspection efficiency and equipment cruising energy are significantly improved.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for locating and targeting damaged netting in deep-sea cages according to the present invention; Figure 2 This is a system module framework diagram of the present invention; Figure 3 This is a flowchart of the wave-tensile amplitude-frequency response operator solution and coherence verification of the present invention; Figure 4 This is a flowchart of the underwater robot visual topology verification process of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0023] Please see Figures 1-4 This invention provides a method for locating and targeting damaged netting in deep-sea cages, comprising the following steps: Step S1: Distribute several tension sensors along the vertical steel ropes of the cage mesh to form a set of measuring points. At each measuring point, a preset sampling rate was used. Simultaneously collect tensile time history data of the netting at each measuring point. Simultaneously, wave measurement equipment was deployed in the waters surrounding the fish cages to acquire wave height time history data at corresponding times. The collected tensile time history data With wave height time history data The original signal is filtered by a bandpass filter to remove high-frequency electrical noise and low-frequency long-period tidal drift, and an effective dynamic time sequence with zero mean is obtained.

[0024] Step S2: Set the time length as The sliding time data window, at any time... Wave height time history signal within a sliding time data window Perform a discrete Fourier transform to obtain the one-sided power spectral density of the wave. Calculate the zero-order moment of the wave based on spectral analysis. With second moment Then, the effective wave height within the window can be obtained. With average zero-crossing cycle .

[0025] Calculate continuous Within a sliding time window Statistical variance and Construct environmental fluctuation discrimination conditions if and only if and (in, , When the preset environmental extreme threshold is reached, the current sea state is determined to be in a wide-stable random process (i.e., a stable sea state period), and frequency domain decoupling calculation of S3 is allowed to be triggered; otherwise, it is determined to be a sea state abrupt change period, and the system automatically filters out false structure alarms caused by the environmental abrupt change.

[0026] Step S3: During periods of stable sea conditions, the system's edge computing nodes retrieve the current... Each duration is Wave height discrete time history signal within the sliding time window With the Discrete time history signal of tension at each measuring point To decouple the random wave excitation in fluid dynamics from the elastic response of the mesh structure, this step uses spectral analysis theory to derive the amplitude-frequency response operator of the structural system, specifically including the following steps: Step S31: Due to the limited length of the actual acquired wave height and tension signal, directly performing a Discrete Fourier Transform would result in severe spectral leakage. The system first performs a zero-homogenization of the discrete signal sequence... and (in , Applying a smoothing Hanning window function to the number of sampling points within the window, the effective signal sequence after windowing is represented as: ; ; Step S32: Since ocean waves are a typical broadband random process, the system uses a modified Welch average periodogram method to estimate the power spectral density of the signal to reduce the variance of the spectral estimation. The windowed signal sequence is overlapped and segmented to calculate the Fourier transform of each segment, and then the averaged wave power spectral density is obtained. The power spectral density of tension The physical meaning of wave self-power spectral density is the wave energy distribution within a unit frequency bandwidth; similarly, the tension self-power spectral density characterizes the frequency domain distribution of tension wave energy, and simultaneously calculates the cross-spectral density between wave excitation and tension response. As shown in the following formula: ; In the formula, For the first Cross-spectral density of a tension sensor The sliding time window step size, The conjugate complex number of the Fourier transform of the wave height signal. For the Fourier transform of the tension signal, This indicates the calculation of the expected value.

[0027] Step S33: Based on the linear system assumption of random vibration theory, the frequency domain response of the mesh structure under wave excitation force can be expressed by the complex transfer function. The system is characterized by solving for the transfer function by the ratio of the cross spectral density to the wave self-power spectral density, as shown in the following equation: ; Amplitude-frequency response operator The physical essence lies in frequency. Under the action of a regular wave, the amplitude of the tension in the steel rope of the netting caused by a unit wave height is determined solely by the inherent stiffness, mass distribution and fluid damping of the netting system, thanks to the effective elimination of interference from the absolute magnitude of the wave height by the amplitude-frequency response operator.

[0028] Step S34: In real deep-sea environments, the data measured by the tension sensor may include not only the response caused by waves, but also nonlinear noise interference such as ocean current impact, large fish school impact, and net vortex-induced vibration. To prevent sudden tension changes caused by non-wave excitation from being misjudged as net damage, the system introduces a coherence coefficient. The validity test is performed as shown in the following formula: ; Set relevant thresholds If in the main energy frequency band of the wave Inside Then the current amplitude-frequency response operator is determined. Valid, proceed to S4.

[0029] Step S4: The essence of mesh damage is stiffness degradation caused by the failure of local structural constraints, which will inevitably be reflected in the distortion of the transfer function. Extract the currently effective transfer function. Main peak frequency within the main energy frequency band of waves Integrate the energy of the curve and compare it with the reference transfer function stored in the system for that measurement point in a healthy state. and its reference main peak frequency Perform comparison, define and calculate measurement points Transfer function distortion rate : ; in, For the normalized weight allocation coefficients, if the distortion rate of the transfer function... Greater than the set safety diagnostic threshold If the test point is detected, it is determined that there is an anomaly in the surrounding netting, triggering a local warning.

[0030] Step S5: Filter out the set of all abnormal measurement points that triggered the warning. Extract the absolute physical coordinates of each abnormal measuring point in the three-dimensional coordinate system of the net cage. and its overlimit weight .

[0031] in, Set a preset safety diagnostic threshold.

[0032] Using a spatial center interpolation algorithm, the three-dimensional target relative coordinates of suspected damaged mesh were calculated. As shown in the following formula: ; in, For all belonging to the set Measurement points in Summation, It is a spatial reduction control factor used to adjust the gravitational effect of high distortion rate measurement points on the positioning center.

[0033] Step S6: Calculate the three-dimensional target coordinates The command is converted into a deep-sea waypoint command that the underwater robot can recognize. The underwater robot only plans the shortest spatial straight line or obstacle avoidance curve path from the current standby origin to the target coordinates.

[0034] The underwater robot reaches the target coordinates and hovers, then activates the computer vision module to acquire local images and perform topological matching. Unlike traditional global edge detection and line fitting, this system uses a graph theory-based local topological feature recognition algorithm to adapt to the flexible deformation of the netting under the action of water flow, including the following steps: Step S61: Due to the scattering and uneven illumination in the underwater optical environment, the vision module first acquires the original video frame, converts it to the LAB color space, retains the luminance channel, and uses a limited contrast adaptive histogram equalization algorithm to enhance local contrast. Then, it uses Gaussian bilateral filtering to smooth high-frequency water suspended noise and acquires a high-contrast net-enhanced image.

[0035] Step S62: The boundary of the mesh is the intersection of multiple mesh feet, which appears in the image as corners with high gradient changes or regions with clustered local high-frequency features. The Harris corner detection operator is used to traverse the pixel matrix of the enhanced image, calculating the local autocorrelation matrix and eigenvalues ​​of each pixel. Pixels with eigenvalues ​​greater than a set threshold are extracted as candidate nodule pixel sets. Since an actual mesh nodule may contain multiple adjacent corner pixels in the image, a density-based spatial clustering algorithm is used to cluster the candidate pixel sets, calculating the geometric center of each cluster and setting its coordinates. Defined as the extracted valid netting knots, constituting a node set. .

[0036] Step S63: Perform adaptive local binarization on the enhanced image to separate the foreground mask of the mesh lines. Use a morphological thinning algorithm to thin the mesh line foreground with a certain width into a skeleton topology map with a single pixel width. Simultaneously, use the node set... Any nodule in Starting from a node, perform a neighborhood search and tracing along a connected single-pixel path on the skeleton topology graph. If the tracing path successfully reaches another node... If the path length is within the set mesh size tolerance range, then it is determined that there is a physical mesh foot between the two points, denoted as an edge. Therefore, all nodes are extracted and an edge set is constructed by iterating through them. .

[0037] Step S64: Based on the extracted nodule set and eye foot set Construct the undirected topology graph of the current local mesh. And generate the corresponding adjacency matrix, from which the undirected topological graph is extracted. Structural characteristics, when the system is in an undirected topological graph When the system detects an abnormal decrease in the degree of multiple consecutive nodes, or when closed loop detection reveals that a closed mesh that should have been composed of 4 edges has become an irregular large hole composed of 6 or even more edges, a rigorous mathematical model can be used to determine that a network break has occurred in the undirected topology graph. The vision module then automatically delineates the irregular topology region on the image with a highlighted bounding box, calculates the actual physical size, and transmits the damage assessment and diagnosis image along with the coordinate data back to the mother ship control center.

[0038] Example 1 like Figure 1 As shown, this embodiment provides a method for locating and targeting damaged netting in deep-sea cages, specifically including the following steps: like Figure 2 As shown, the physical structure parameters of the cage are input into the system through the data synchronization acquisition module. The parameters are evenly distributed along the vertical main steel ropes of the cage's mesh according to the spatial topology. A set of miniature tensile sensors constitutes the measuring points. Set the sampling frequency. The frequency is 10Hz, and the discrete time history signal of the tension of the netting at each measuring point is collected synchronously. Simultaneously, wave-measuring radars were deployed around the wave-facing surface of the area where the net cages were moored, using the same sampling frequency. Synchronous acquisition of sea surface time-history discrete signals After receiving the signal, the system uses a bandpass filter to remove high-frequency electrical noise and low-frequency tidal drift caused by sensor zero-point drift, obtains the effective dynamic time sequence with zero mean, and stores it in the data buffer pool.

[0039] To filter out the interference of extreme marine weather changes on structural response assessment, the edge computing module performs sea state stability verification before frequency domain calculation, with a set time length of [missing information]. A sliding data window. In the... Within a sliding window, the wave height time history signal Perform a fast Fourier transform to obtain the one-sided power spectral density of the wave. Calculate the zeroth moment of the spectrum. and second moment As shown in the following formula: ; And derive the effective wave height within this window. With average zero-crossing cycle : ; ; System continuous statistics Feature pairs within a sliding time window Time series variance and If and only if and (in, , When the preset environmental fluctuation extreme threshold is reached, the current sea state is determined to be in a wide-stable random process, triggering the frequency domain calculation of S3; otherwise, the system suspends feature extraction and enters the next sliding window to filter out false alarms caused by sudden environmental changes.

[0040] During periods of calm sea conditions, for the first Wave height signal within a sliding window With the Tension signal at each measuring point A Hanning window is applied for truncation; the window function expression is shown below: ; The wave self-power density was obtained using the modified Welch average periodogram method. Tensile power spectral density and their cross-spectral density As shown in the following formula: ; Based on the linear system assumption in random vibration theory, the derivation of wave excitation to the measuring point is performed. Complex transfer function of tensile response And extract its modulus as the amplitude-frequency response operator. Amplitude-frequency response operator The expression is shown in the following formula: ; Simultaneous calculation of coherence function As shown in the following formula: ; Set a coherence threshold Only when the main energy frequency band of the wave... Inside If the response is deemed valid, the damage diagnosis process can proceed.

[0041] Extracting the valid transfer function Main peak frequency within the main frequency band Integrate the curve energy and compare it with the pre-stored healthy baseline parameters. Perform a comparison. Calculate the measurement points. Transfer function distortion rate As shown in the following formula: ; in These are the normalized weighting coefficients. If This triggered a local early warning.

[0042] Extract the set of abnormal measurement points that exceed the limit Read the three-dimensional physical coordinates of each abnormal measurement point in the global coordinate system of the net cage. and its overlimit weight .

[0043] Using an inverse distance spatial weighted algorithm, the three-dimensional target relative coordinates of suspected damaged mesh were calculated. The expression for the relative coordinates of the three-dimensional target is shown in the following formula: ; The expression is as follows: ; ; ; in, The absolute physical coordinates of the abnormal measurement points. The excessive weight of this abnormal measurement point. This is the spatial decay control factor.

[0044] The underwater robot received Following the command, the shortest descent path is planned using an ultra-short baseline positioning system. Upon arrival, the set of netting nodules is extracted using a computer vision module. Construct a local undirected topological graph with the connected target set E. When the algorithm detects an abnormal decrease in the degree of multiple consecutive internal nodes in G, it diagnoses a break in the network cable at that location.

[0045] The following calculation example illustrates the calculations. Taking a deep-sea gravity-type cylindrical net cage with a diameter of 30m and a water depth of 15m as an example, 12 vertical main steel cables are installed around the perimeter of the net (the angle between adjacent cables is...). The netting material is ultra-high molecular weight polyethylene (UHMPWE), with a standard rhomboid mesh shape and a mesh size (half-mesh length) of 2.5 cm. High-frequency miniature tension sensors are connected in series on each steel cable at depths of -5m, -10m, and -15m, forming a tension sensor array with a total of 36 measuring points. Simultaneously, a wave monitoring buoy equipped with a built-in gyroscope and accelerometer is deployed 100m directly in front of the net cage's wave-facing side. Both the tension sensor array and the wave buoy synchronously acquire data at a sampling frequency of 10Hz, and the data is aggregated in real-time to the edge computing node of the central control platform at the sea surface via an underwater acoustic communication relay node.

[0046] Edge computing nodes are equipped with a Python computing environment, primarily relying on underlying data analysis and signal processing frameworks for real-time stream computing. A sliding time window is set. (That is, 10800s, containing 108000 discrete sampling points). During a certain test period, the wave buoy transmits irregular wave height time series data. The system Perform a Fast Fourier Transform (FFT) to solve the one-sided power spectrum and calculate its zeroth moment. With second moment The effective wave height within the current sliding window has been calculated. Average cross-zero cycle .

[0047] The system traces back four past sliding windows (total) The variance of the significant wave height was calculated from the sea state characteristic parameters within the range. The system's preset environmental fluctuation extreme value threshold .because The system determines that no sudden changes such as storm surges have occurred in the current sea area, and that it is in a period of wide-range stable random sea state, which meets the prerequisite for frequency domain decoupling, and automatically triggers the core transfer function calculation module.

[0048] like Figure 3 As shown, the water depth on the main rope at the wave-facing side is... The measuring point at the location (set as sensor) Its physical coordinates in the Cartesian coordinate system with the center of the cage as the origin are: Let's take ) as an example to make a deduction.

[0049] The edge computing node's waveform height signal within the window With tension signal The signal was truncated by applying a Hanning window, and the cross spectral density and wave self-power spectral density were calculated using the Welch average periodogram method. According to the formula Derivation of the sensor The amplitude-frequency response operator at the current location. Simultaneously, to visually record the health status of the mesh, the system calls the matplotlib plotting library module to automatically generate and save the amplitude-frequency response curve and coherence function curve for the current time window to the system log.

[0050] The system synchronously calculates the coherence function at this measuring point. In areas where wave energy is concentrated Within the main frequency band, calculations yielded The average value is .because (Preset coherence threshold) The system determines that the tension fluctuation is dominated by linear wave excitation and is not affected by nonlinear noise pollution such as impacts from large marine organisms, and the current transfer function data is valid.

[0051] The system extracts the current The main peak frequency of the curve is identified as follows: The system reads the initial health baseline parameters of the net cage from its internal database to obtain the data from the measuring points. The main peak frequency under normal conditions should be Substitute into the distortion rate calculation formula (set normalization weighting coefficients). After combining the main frequency offset and the energy envelope difference, the current measurement point is calculated. Transfer function distortion rate .because (Preset safety diagnostic threshold) =15), the system determined that the stiffness of the mesh near the measuring point had severely degraded, triggering a high-priority local damage warning. Through synchronous scanning, adjacent sensors were detected. It also exceeds the limit, and its distortion rate .

[0052] This embodiment uses the two adjacent tension sensors mentioned above. , The spatial location data is used to solve the problem. Spatial coordinates are , , , Spatial coordinates are , , And excessive weight , .

[0053] The calculation results based on the expression for the relative coordinates of the three-dimensional target are as follows: ; ; ; Thus, the system successfully converted macroscopic sensor warning signals into highly directional spatial physical coordinates. , , ).

[0054] like Figure 4 As shown, the central control platform sends the aforementioned target coordinates to the underwater robot (ROV) that is parked and waiting underwater. The RO main control unit parses the sent command packet, calls the built-in ultra-short baseline positioning system (USBL) and inertial navigation system (INS), and calculates the current standby position. The spatial motion vector.

[0055] The ROV's thruster control algorithm (such as incremental PID control) drives the device to descend along the shortest safe straight path. It reaches a preset safe distance directly in front of the target coordinates (e.g., ...). After that, the thruster enters a constant depth hovering mode. The ambient light sensor detects the current water depth illuminance. If it is below the threshold, the ROV automatically activates the adaptive LED supplemental lighting matrix to eliminate underwater shadows and compensate for color attenuation.

[0056] The ROV is equipped with a high-definition underwater camera that sets the frame rate (e.g., The system captures the dynamic video stream of the mesh fabric. The vision processing unit calls OpenCV's cv2.VideoCapture interface to read the video stream, extracts consecutive video frames frame by frame according to timestamps, decodes them losslessly into independent BMP image sequences, and stores them in a memory buffer to avoid video compression artifacts interfering with subsequent topology recognition.

[0057] For a single-frame BMP image, `cv2.cvtColor` is called to convert it from the BGR color space to a grayscale single-channel image. To overcome uneven underwater illumination and scattering by suspended particles, `cv2.createCLAHE` is called to apply the Limiting Contrast Adaptive Histogram Equalization (CLAHE) algorithm to stretch the contrast of local mesh areas. `cv2.bilateralFilter` is then called for bilateral filtering and smoothing. This operation effectively filters out high-frequency suspended particle noise in the water while strictly preserving the sharpness information of the mesh edges.

[0058] Meanwhile, mesh nodes appear as local high-gradient extrema points where multiple line segments intersect in the image. This necessitates calling OpenCV's Shi-Tomasi corner detection operator `cv2.goodFeaturesToTrack` to calculate the pixel feature matrix on the enhanced grayscale image. The system extracts a set of strong corner points based on a preset quality threshold (e.g., 0.01) and a minimum Euclidean distance (e.g., 10 pixels). Considering that a physical nodule may be detected as several adjacent corner points in the image, the DBSCAN clustering algorithm is used to perform secondary spatial clustering on the aforementioned corner point set. The geometric moment center coordinates of each valid cluster are calculated. This allows for precise identification of the nodes as "knots" within the mesh, generating a node set. This precise labeling of feature points laid the foundation for subsequent topology map construction.

[0059] However, under the impact of water flow, the mesh lines (eyelets) are not rigid straight lines, but rather exhibit a flexible, curved shape. Therefore, adaptive binarization is required. The `cv2.adaptiveThreshold` algorithm is called to segment the grayscale image into a binary foreground (mesh lines) and background (water) based on the average pixel value of the local neighborhood. Simultaneously, for the wide-width binary mesh foreground, a thinning algorithm (such as the Zhang-Suen algorithm implemented in `cv2.ximgproc.thinning`) is called to continuously peel away the foreground boundary pixels until a single-pixel-width mesh skeleton topology is extracted. (Based on a node set...) nodes in Starting from the white (value 255) pixel, perform an eight-neighbor search along the white pixel in the skeleton topology graph. If the search path leads to another node... If the path length is within the normal half-eye length tolerance range, then a physical "eye foot" is confirmed to exist and is recorded as an undirected edge. Iterate through all nodes and generate a set of connected edges. .

[0060] Finally, the adjacency matrix is ​​constructed based on the extracted node set. With edge set Construct an undirected graph topology model of the current local net in memory. This process generates an adjacency matrix representing the connectivity relationships between nodes. The topology graph is then traversed. The degree of each node (i.e., the number of nodes connected to that node). In a standard diamond mesh, the degree of an intact node is always 4. The system algorithm sets the diagnostic logic: if the degree of consecutive adjacent nodes abnormally drops to 2 or 3, or if an abnormally enlarging closed loop consisting of 6 or more edges is detected by depth-first search (DFS), then a physical break in the mesh is diagnosed based on a rigorous mathematical model.

[0061] After damage is confirmed, `cv2.drawContours` and `cv2.rectangle` are used to automatically delineate the abnormal topology region on the original BMP image using a highlighted red bounding box. Finally, `cv2.VideoWriter` is used to re-synthesize the processed BMP image sequence with highlighted warning boxes and damage assessment size marks into an early warning review video stream, which, along with the final confirmation report, is uploaded to the mother ship control center via underwater acoustic communication or umbilical cable, completing the full closed loop of targeted inspection.

[0062] Therefore, the present invention adopts the above-mentioned method and system for locating and targeting damaged netting in deep-sea cages, which utilizes frequency domain characteristics to reveal structural stiffness degradation, effectively filters out random interference from sea conditions, and realizes a closed loop of electromechanical collaborative targeted inspection with high anti-interference, low computing power consumption and high efficiency.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for locating and targeting damaged netting in deep-sea cages, characterized in that, Includes the following steps: S1. Tension sensors are installed on the netting of the cage to collect tension time history data synchronously, and wave measurement equipment is installed around the cage to acquire wave height time history data synchronously. S2. Set the sliding time data window, calculate the sea state characteristic parameters based on the wave height time history data and calculate their variance. When the variance meets the preset environmental fluctuation discrimination conditions, it is determined that the sea state is in a stable period and the frequency domain decoupling calculation is triggered. Otherwise, it is determined to be a sea state change period and the alarm is filtered out. S3. During the period of stable sea conditions, the amplitude-frequency response operator between wave excitation and net tension response is calculated using spectral analysis theory, and a coherence function is introduced to test its effectiveness. If the coherence function is greater than the set threshold, the current amplitude-frequency response operator is determined to be effective. S4. Extract the main peak frequency and curve energy integral of the effective amplitude-frequency response operator in the main energy frequency band of the wave, and compare it with the pre-stored health benchmark transfer function. Calculate the transfer function distortion rate. If the distortion rate is greater than the safety diagnosis threshold, trigger a local early warning. S5. Extract the absolute physical coordinates and over-limit weights of the abnormal measuring points that trigger local early warning, and calculate the three-dimensional target relative coordinates of the suspected damaged mesh. S6. Send the three-dimensional target relative coordinates to the underwater robot. The underwater robot autonomously navigates to the three-dimensional target relative coordinates and starts the computer vision module to extract the set of knots and feet of the netting to construct an undirected topology graph. By judging the abnormality of the structural features of the topology graph, the physical breakage of the netting is diagnosed, and the damage assessment diagnosis image and coordinate data are transmitted back.

2. The method for locating and targeting damaged netting in deep-sea cages according to claim 1, characterized in that, S1 includes: S11. Set up several measuring points on the vertical steel ropes of the cage mesh, and arrange several tension sensors at the measuring points to form a set of measuring points. ; S12. At each measuring point, use a preset sampling rate. Synchronous data acquisition was performed to obtain the tensile time history data of the netting at each measuring point. ; S13. Deploy wave measurement equipment in the sea area surrounding the net cages to simultaneously acquire wave height time history data at corresponding times. ; S14. Collect the tensile time history data. With wave height time history data The original signal is filtered by a bandpass filter to remove high-frequency electrical noise and low-frequency long-period tidal drift, and an effective dynamic time sequence with zero mean is obtained.

3. The method for locating and targeting damage to deep-sea cage mesh as described in claim 2, characterized in that, The process for determining that the sea state is in a stable period in S2 is as follows: S21, in any... Within a sliding time data window, the wave height time history signal is subjected to a discrete Fourier transform to obtain the wave's one-sided power spectral density, and the zero-order moment is calculated. With second moment ; S22. Calculate the effective wave height within this window. With average zero-crossing cycle ; S23, Statistical Continuity The variance of the effective wave height and the average zero-crossing period within each sliding time window and If and only if and At that time, it was determined that the sea state was in a period of stability, among which This is a preset environmental extreme value threshold.

4. The method for locating and targeting damage to deep-sea cage mesh as described in claim 3, characterized in that, The process of calculating the amplitude-frequency response operator and performing message verification in S3 is as follows: S31. Apply a smoothing Hanning window function to the zero-mean-normalized wave height discrete signal sequence and the tension discrete signal sequence. S32. Calculate the averaged wave power spectral density using the Welch average periodogram method. Tensile power spectral density and cross spectral density ; S33. Solve the complex transfer function by the ratio of the cross spectral density to the wave self-power spectral density, and extract its magnitude as the amplitude-frequency response operator. ; S34. Calculate the coherence function If within the main energy frequency band of the wave If so, the current amplitude-frequency response operator is considered valid, where, This is the coherence threshold.

5. The method for locating and targeting damage to deep-sea cage mesh as described in claim 4, characterized in that, The coherence function in S34 As shown in the following formula: 。 6. The method for locating and targeting damage to deep-sea cage mesh as described in claim 5, characterized in that, In S4, the transfer function distortion rate is calculated using the following formula. : ； in, and These are the peak frequency and amplitude of the current effective transfer function, respectively; and These represent the main peak frequency and amplitude of the reference transfer function, respectively. These are the normalized weighting coefficients.

7. The method for locating and targeting damaged netting in deep-sea cages according to claim 6, characterized in that, In S5, the three-dimensional target relative coordinates of the suspected damaged mesh are calculated using the following formula. : ； ； ； in, The absolute physical coordinates of the abnormal measurement points. The excessive weight of this abnormal measurement point. This is the spatial decay control factor.

8. The method for locating and targeting damage to deep-sea cage mesh as described in claim 7, characterized in that, The graph-based local topological feature recognition algorithm in S6 extracts the set of nodes of the mesh, including: The original video frames are acquired and converted to the LAB color space, the luminance channel is preserved, the local contrast is enhanced by the contrast-limited adaptive histogram equalization algorithm, and the high-frequency noise is smoothed by Gaussian bilateral filtering to obtain the enhanced image. The Harris corner detection operator is used to traverse and enhance the image to calculate feature values, and pixels with feature values ​​greater than a set threshold are extracted as candidate nodule pixel sets. A density-based spatial clustering algorithm is used to cluster the candidate nodule pixel set. The geometric center of each cluster is calculated as the effective netting nodule, forming a node set. .

9. A method for locating and targeting damaged netting in deep-sea cages according to claim 8, characterized in that, The graph theory-based local topological feature recognition algorithm in S6 aims to construct an undirected topological graph from a set of footprints, including: Adaptive local binarization is applied to the enhanced image to separate the foreground mask, and a morphological thinning algorithm is used to refine the mesh foreground into a skeleton topology map with a single pixel width. With node set Starting from any node in the graph, perform an eight-neighbor search along a connected single-pixel path on the skeleton topology graph. If the tracing path successfully reaches another node and its length is within the tolerance range, then it is determined that there is a physical destination between the two points, forming an edge set. ; Based on node set Sum of edges Constructing an undirected topological graph of the local mesh When an abnormal decrease in the degree of multiple consecutive nodes is detected in an undirected topology graph, or when the number of edges constituting a closed mesh is found to have increased to the threshold of an irregular polygon through closed loop detection, it is determined that a network break has occurred in the undirected topology graph.

10. A system applied to the deep-sea cage mesh damage location and targeted inspection method described in claim 9, characterized in that, include: The data synchronization acquisition module is used to synchronously acquire the tension time history data and the wave height time history data of the netting at each measuring point, and perform filtering preprocessing. The edge computing and frequency domain decoupling module is used to perform sea state stability verification and calculate the amplitude-frequency response operator, coherence function and transfer function distortion rate during the sea state stability period, thereby triggering local early warning; The spatial positioning and command generation module is used to extract the absolute physical coordinates and their over-limit weights of abnormal measurement points, calculate the three-dimensional target relative coordinates of suspected damaged mesh, and generate waypoint commands for the underwater robot. The underwater robot execution module is used to autonomously navigate to the target coordinates after receiving instructions. It uses a computer vision module to extract the knot set of the netting and construct an undirected topology graph with the target feet. It diagnoses physical breaks in the netting through topological anomalies and transmits damage assessment data back.

Images