Three-dimensional modeling method and system for underwater survey and site selection of artificial fish reefs

Abstract

The application discloses a three-dimensional modeling method and system for underwater survey and site selection of artificial reefs, and particularly relates to the technical field of underwater survey data processing, which comprises obtaining multi-source survey data of a candidate site selection area, aligning each data to the same coordinate system according to a track time scale by an edge computing node, and outputting a registration data set. The method is based on the edge computing node, completes multi-source survey data registration according to the track time scale, extracts four types of risk boundary, such as a sonar shadow boundary, to cut risk voxels, writes data in the slot according to the evidence level in the voxel, and executes a closed judgment, finally generates a three-dimensional model with an evidence state, and outputs a targeted supplementary survey task, which can avoid the misjudgment of site selection with a complete surface but insufficient evidence, and improve the efficiency of underwater survey supplementary survey.

Description

Technical Field

[0001] This invention relates to the field of underwater survey data processing technology, and more specifically, to a three-dimensional modeling method and system for underwater site selection of artificial reefs. Background Technology

[0002] In the field of underwater surveying and site selection for artificial reefs, the mainstream practice in the industry is to use multibeam sonar, side-scan sonar, underwater vision, inertial navigation and depth sensors to conduct aerial surveys of candidate sea areas, and then use shore-based servers or shipborne computing platforms to fuse point cloud, image and pose data to obtain a three-dimensional model of the seabed topography, and further determine the placement location of artificial reefs based on water depth, slope, seabed sediment and obstacle distribution.

[0003] In actual operations in turbid nearshore waters, AUVs or ROVs often need to complete candidate area surveys under hard constraints such as low-bandwidth underwater acoustic communication, limited battery capacity, continuous attitude disturbances, suspended sediment obstruction, and the inability to transmit full amounts of raw data for extended periods. Therefore, although edge computing can be used for on-site data compression, preliminary filtering, or generation of local 3D models, existing methods still mainly rely on point cloud continuity, mesh integrity, or sensor coverage as the basis for model completion. This results in the model surface generated at the edge end appearing as continuous, smooth, and fully observable results in areas such as local slope toe, soft and hard seabed boundaries, sonar shadow areas, or visually turbid areas. However, the actual seabed stability, placement attitude risk, hydrodynamic disturbances, and missing measurement evidence used to support artificial reef site selection are not closed. Ultimately, the model is completed, but the reliability of the candidate reef site still needs to be manually verified after returning to shore. In some cases, the modeled area is re-surveyed or the site selection conclusion is overturned.

[0004] The technical problem this application aims to solve is: how to overcome the limitations of existing 3D modeling which only uses geometric integrity as the criterion in an edge computing environment, to achieve closed-loop determination of evidence related to artificial reef site selection, and to avoid generating 3D models with complete surfaces but insufficient decision-making basis; and how to screen the missing measurement areas that truly affect the site selection conclusion based on the evidence closure status, improve the pertinence of on-site supplementary measurements, and reduce invalid survey operations. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, this invention provides a three-dimensional modeling method and system for underwater surveying and site selection of artificial reefs. This method can achieve closed-loop determination of site selection evidence under edge computing conditions, avoiding the generation of three-dimensional models with complete surface geometry but insufficient to support deployment decisions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a three-dimensional modeling method for underwater surveying and site selection of artificial reefs, comprising: S1. Obtain multi-source survey data of the candidate site selection area, and have the edge computing nodes align each data to the same coordinate system according to the track time scale, and output the registration dataset; S2. Input the registration dataset into the edge computing node. The edge computing node extracts the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment. Using the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment as voxel boundaries, the candidate site selection area is segmented and the risk voxel set is output. S3. For each risk voxel, establish a topographic trough, sediment trough, flow trough, and landing trough, and write the registration data falling into the risk voxel into the corresponding trough. The original measured data is written directly. When two data from different sources point to the same trough result, they are merged and written. Only the result obtained from adjacent risk voxels is downgraded and written. When data from different sources point to opposite trough results, a conflict marker is written. Output a set of risk voxels with trough markers. S4. Perform closure determination on the risk voxel set with slot markings: When there are direct write or merged write results in the topographic slot, bottom sediment slot, flow slot and landing slot, and there are no conflict markings in the four slots, the corresponding risk voxel is recorded as a closed voxel; otherwise, it is recorded as an unclosed voxel, and the missing certificate slot is recorded. The closed voxel, unclosed voxel and missing certificate slot are output. S5. Generate a 3D model based on closed voxels, unclosed voxels, and missing evidence slots. Then, perform a site selection judgment for each unclosed voxel after replacing it with the favorable and unfavorable results of the missing evidence slot. If the two judgment results are different, list the unclosed voxel as a supplementary test object and output the evidence closure type 3D model and the supplementary test task.

[0007] In a preferred embodiment, S1 includes: S1-1. Input the multi-source survey data of the candidate site selection area into the edge computing node. The edge computing node uses the track time of the underwater survey carrier as the main sequence, writes the acquisition time of each survey data between the two adjacent track points, and outputs the survey dataset with track interval markings. S1-2. Input the survey dataset with track interval markers into the edge computing node. The edge computing node calculates the registration pose of the survey data according to the time position of each survey data in the corresponding track interval, along the pose change direction of the track points at both ends of the track interval, and projects the survey data to the same coordinate system according to the registration pose, and outputs the initial registration dataset. S1-3. Input the initial registration dataset into the edge computing node. The edge computing node retains the original source, track interval label and registration pose of the survey data projected to the same spatial location but with different track interval labels. It does not merge the survey data into a single result and outputs the registration dataset for subsequent risk voxel writing.

[0008] In a preferred embodiment, S2 includes: S2-1. Input the registration dataset into the edge computing node. The edge computing node reads the sonar echo missing area, seabed elevation turning area, seabed sediment type switching area and pose change area along the track direction of the underwater survey vehicle. It then extracts the corresponding sonar shadow boundary, slope line, seabed sediment change line and pose change segment, and outputs the risk boundary set. S2-2. Input the risk boundary set into the edge computing node. The edge computing node connects the intersecting or adjacent sonar shadow boundaries, slope toe lines, bottom change lines and attitude change segments into closed or semi-closed dividing lines. The unclosed dividing lines are extended to the boundary of the candidate site selection area, and the risk dividing line set is output. S2-3. Input the risk segmentation line set into the edge computing node. The edge computing node uses the risk segmentation line set as the voxel segmentation boundary to perform non-uniform spatial segmentation of the candidate site selection area, so that the same risk voxel does not cross the sonar shadow boundary, slope toe line, bottom change line or attitude change segment, and outputs the risk voxel set.

[0009] In a preferred embodiment, S3 includes: S3-1. Input the registration data falling into the same risk voxel into the edge computing node. The edge computing node constructs the observation matrix according to the topographic trough, bottom sediment trough, flow trough and location trough respectively. Then, it performs singular value decomposition on the observation matrix to retain the measured components corresponding to the main singular direction and remove the drift components orthogonal to the main singular direction. The direct observation vector of each trough is output. S3-2. Input the direct observation vector into the edge computing node. The edge computing node uses a regularized linear solution algorithm to solve the candidate writing results of each slot, constrained by the consistency of slot results, the continuous change of adjacent risk voxels, and the limitation of residuals from the same source. It outputs the candidate results and residual records of each risk voxel in the topographic slot, bottom sediment slot, water flow slot and landing slot.

[0010] In a preferred embodiment, S3 further includes: S3-3. Input the candidate results and residual records into the edge computing node. The edge computing node compares the directions of candidate results from different sources in the same slot. If the directions are consistent, they are merged and written while retaining the source mark. If the candidate result is only obtained by the regularization term of the adjacent risk voxel, it is downgraded and written as a transmission result. If the directions are opposite, merging is not performed and a conflict mark is written. The slot result with source mark, downgrade mark and conflict mark is output. S3-4. Write the slot results with source markers, degradation markers and conflict markers back to the corresponding risk voxels. The edge computing nodes prohibit degradation markers from replacing direct observation results alone and prohibit conflict markers from being covered by regularization results. Output the risk voxel set with slot markers.

[0011] In a preferred embodiment, S4 includes: S4-1. Input the risk voxel set with slot markings into the edge computing node. The edge computing node reads the topographic slot, bottom sediment slot, water flow slot and landing slot of each risk voxel one by one. The direct write results and merged write results in each slot are recorded as valid evidence. The downgraded write results, empty slots and conflict markers are recorded as non-closed evidence. The slot evidence table of each risk voxel is output.

[0012] In a preferred embodiment, S4 further includes: S4-2. Input the slot evidence table into the edge computing node. The edge computing node performs closure verification in a fixed order of topographic slot, bottom sediment slot, water flow slot and landing slot. If all four slots are valid evidence and there are no conflict markers, the corresponding risk voxel is recorded as a closed voxel. If any slot is non-closed evidence, the corresponding risk voxel is recorded as an unclosed voxel and the slot is extracted as a missing evidence slot. The initial closure judgment result is output.

[0013] S4-3. Input the initial closure judgment result into the edge computing node. The edge computing node connects and merges the missing certificate slots of adjacent risk voxels, and merges unclosed voxels with the same missing certificate slots and spatially adjacent voxels into the same missing certificate block. Output closed voxels, unclosed voxels and missing certificate slots.

[0014] In a preferred embodiment, S5 includes: S5-1. Input closed voxels, unclosed voxels, and missing voxels into the edge computing node. The edge computing node will build a voxel map with each voxel as a node and coplanar adjacent voxels as connecting edges. The topographic voxel, bottom sediment voxel, water flow voxel, and landing voxel of each voxel will be converted into the voxel position state, and the voxel position state map will be output.

[0015] S5-2. Input the slot status diagram into the edge computing node. The edge computing node generates three-dimensional model elements according to the voxel space boundary. The coplanar adjacent three-dimensional model elements are merged into their common surface. The slot status of the corresponding voxel is written into the corresponding three-dimensional model element, and the evidence closed three-dimensional model is output.

[0016] S5-3. Input the evidence-closed 3D model into the edge computing node, and have the edge computing node copy the local model containing unclosed voxels one by one; write favorable and unfavorable results for the missing evidence slots of the same unclosed voxel respectively. The favorable result for the topographic slot is a continuous support surface, and the unfavorable result is a slope toe fracture. The favorable result for the bottom slot is a hard bottom, and the unfavorable result is soft mud. The favorable result for the water flow slot is bypass flow, and the unfavorable result is direct flow or back siltation. The favorable result for the placement slot is complete bottom surface contact, and the unfavorable result is bottom surface suspension or obstacle occupation. Output favorable and unfavorable replacement models.

[0017] In a preferred embodiment, S5 further includes: S5-4. Input the favorable replacement model and unfavorable replacement model into the edge computing node. The edge computing node reads the three-dimensional model units in the same candidate placement area according to the occupied range of the artificial reef bottom surface. When all the read three-dimensional model units have favorable results and no conflict markers, they are judged as selectable. When any three-dimensional model unit has unfavorable results, missing slots, or conflict markers, it is judged as unselectable. Output the two site selection judgment results.

[0018] S5-5. Input the results of the two site selection judgments into the edge computing node. When the edge computing node determines that one candidate site is selectable and the other is not selectable in the same candidate site area, list the corresponding unclosed voxels as supplementary measurement objects and generate supplementary measurement tasks according to their missing evidence slots. Among them, the topographic slot corresponds to sonar rescanning, the sediment slot corresponds to bottom surface resampling, the water flow slot corresponds to flow direction rescanning, and the site slot corresponds to surround rescanning. Output the evidence closure type 3D model and supplementary measurement tasks.

[0019] In a preferred embodiment, the 3D modeling system for underwater surveying and site selection of artificial reefs includes a registration and assembly module, a segmentation module, a trench marking module, a closure and missing information identification module, and a modeling and supplementary survey module. The registration and set module is used to acquire multi-source survey data of candidate site selection areas, and the edge computing nodes align each data to the same coordinate system according to the track time scale, and output the registration dataset.

[0020] The voxel segmentation module is used to input the registration dataset into the edge computing node. The edge computing node extracts the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment. Using the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment as voxel boundaries, the candidate site selection area is segmented and the risk voxel set is output.

[0021] The slot labeling module is used to create topographic slots, sediment slots, flow slots, and landing slots for each risk voxel, and write the registration data falling into the risk voxel into the corresponding slot. The original measured data is written directly, when two data from different sources point to the same slot result, they are merged and written, only the result obtained from adjacent risk voxels is downgraded and written, and when data from different sources point to opposite slot results, a conflict marker is written, and a set of risk voxels with slot markers is output.

[0022] The closure and deficiency identification module is used to perform closure determination on risk voxel sets with slot markings: when there are direct write or merged write results in the topographic slot, bottom sediment slot, water flow slot and landing slot, and there are no conflict markings in the four slots, the corresponding risk voxel is recorded as a closed voxel; otherwise, it is recorded as an unclosed voxel, and the missing certificate slot is recorded, and the closed voxel, unclosed voxel and missing certificate slot are output.

[0023] The modeling and supplementary testing module is used to generate a 3D model based on closed voxels, unclosed voxels, and missing evidence slots. For each unclosed voxel, it performs a site selection judgment after replacing it with the favorable and unfavorable results of the missing evidence slot. If the two judgment results are different, the unclosed voxel is listed as a supplementary testing object, and the evidence closure type 3D model and supplementary testing task are output.

[0024] The beneficial effects of this invention are: By setting terrain troughs, substrate troughs, water flow troughs, and location troughs for risk voxels in edge computing nodes, and replacing simple geometric integrity judgment with closure judgment, the 3D model can simultaneously reflect whether the site selection evidence is sufficient, thereby relatively reducing the risk of misjudgment due to surface integrity but insufficient decision evidence.

[0025] By registering multi-source data according to track time scales and retaining the source, interval and pose information of different track intervals, repeated observations at the same spatial location are not averaged in advance, which relatively improves the ability to identify conflicts caused by attitude disturbances and repeated aerial scans. By using sonar shadow boundaries, slope toe lines, substrate change lines, and attitude change segments as voxel boundaries, risk voxels are prevented from crossing key risk locations, thus relatively reducing the occurrence of opposite terrain or substrate evidence being written into the same spatial cell.

[0026] By using direct write, merge write, downgrade write, and conflict-marked slot write rules, measured evidence, transmitted results, and opposite results are processed separately, which relatively suppresses the problem of interpolation or smoothing results masking the missing measurement areas.

[0027] By extracting the main measured components within the slot using singular value decomposition in the edge computing node, and combining this with regularized linear solution to generate candidate write results, the edge end retains the main observation information under limited computing power, and relatively reduces the interference of attitude drift on the slot results.

[0028] By substituting favorable and unfavorable results into the unclosed voxels to perform site selection determination, only voxels that will reverse the site selection conclusion will be listed as supplementary testing targets. This can relatively reduce invalid supplementary testing and improve the targeting of on-site supplementary testing tasks. Attached Figure Description

[0029] Figure 1 This is a flowchart of the present invention.

[0030] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Refer to the instruction manual appendix Figure 1-2 The present invention provides a three-dimensional modeling method for underwater surveying and site selection of artificial reefs, comprising: S1. Obtain multi-source survey data of the candidate site selection area, and have the edge computing nodes align each data to the same coordinate system according to the track time scale, and output the registration dataset; To enable survey data from different sources and with different acquisition frequencies to be included in subsequent risk voxel writing, this embodiment uses the track timescale of the underwater survey vehicle as the main timeline. First, the track interval to which each survey data belongs is determined, and then the registration pose is calculated according to its time position in the track interval. The data is then projected onto the same coordinate system. For data that fall in the same spatial position after projection but come from different track intervals, they are not directly merged in the registration stage. Instead, the source, track interval, and registration pose are retained for subsequent determination of whether they belong to consistent evidence, conflicting evidence, or sources of missing evidence. This implementation process includes the following steps.

[0033] In S1-1, after the multi-source survey data of the candidate site selection area is input into the edge computing node, the edge computing node reads the track point sequence of the underwater survey vehicle. Each track point includes the acquisition time, location, and attitude, and is arranged according to the acquisition time to form a track time stamp. Then, the acquisition time of each survey data is read, and the nearest track point before and after the acquisition time is found in the track time stamp. The survey data is assigned to the track interval formed by these two track points, and the original source, acquisition time, and track interval label are recorded. If the acquisition time of the survey data is the same as a certain track point, the data is assigned to the adjacent track interval with that track point as the endpoint according to the vehicle's navigation direction. Thus, a survey dataset with track interval labels is output. The purpose of this processing is to make each survey data correspond to a segment of vehicle movement process, rather than relying on only a single time point. For example, image frames, sonar echoes, and water flow records have different sampling frequencies. If they are not assigned to track intervals first, it will be difficult to determine whether they correspond to the same segment of observation or repeated observations of different segments.

[0034] In S1-2, after the survey dataset with track interval markers is input into the edge computing node, the edge computing node reads the two endpoint track points of the track interval to which each survey data belongs, and calculates the time position of the survey data within the track interval based on the position of the survey data acquisition time between the start and end times of the track interval. Subsequently, the registration pose of the survey data is calculated along the position and attitude changes of the two endpoint track points, and the survey data is projected from the sensor coordinates or carrier coordinates to the same coordinate system of the candidate site selection area based on the registration pose to obtain the initial registration dataset. Specifically, if a sonar data is acquired between two track points, its spatial position is not directly taken from the previous or next track point, but is calculated according to its position within the track interval at the time of acquisition. This can reduce the projection deviation caused by carrier movement and avoid mistakenly overlaying the data acquired during navigation onto a certain track point.

[0035] In S1-3, after the initial registration dataset is input into the edge computing node, the edge computing node performs spatial aggregation on the projected data. If multiple survey data fall into the same subsequent risk voxel, the same spatial cell, or the same spatial grid to be modeled, they are considered to have been projected to the same spatial location, and their track interval labels are compared. If the track interval labels are the same, they are treated as data of the same flight segment when writing to the subsequent slots. If the track interval labels are different, their original sources, track interval labels, and registration poses are retained respectively, and they are not merged into a single result during the registration stage. The registration dataset used for subsequent risk voxel writing is output. The reason for this processing is that data from different flight segments at the same location may come from different perspectives or different attitudes. Direct merging will smooth out the real conflicts. For example, the same slope foot position may show continuous echoes in one aerial scan and missing echoes in another aerial scan. If it is directly averaged at this time, it will weaken the basis for judging the subsequent sonar shadow boundary and conflict marker.

[0036] Through the above implementation process, the edge computing node completes track interval marking, registration pose calculation, and coordinate projection on-site, while preserving the data differences when different track intervals fall into the same spatial location. This allows subsequent risk voxel writing to distinguish between direct observation, repeated observation, and potential conflicts. In practical applications: when an underwater survey vehicle is conducting aerial scanning along a candidate site, the sonar data of the first segment and the image data of the second segment may fall at the same toe position. The edge computing node records the source, track interval, and registration pose of both. When writing risk voxels, if both point to a continuous toe, they can be used as evidence of consistency. If one points to a continuous support surface and the other points to a fracture at the toe, they are not merged in advance but enter the conflict writing process.

[0037] S2. Input the registration dataset into the edge computing node. The edge computing node extracts the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment. Using the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment as voxel boundaries, the candidate site selection area is segmented and the risk voxel set is output.

[0038] To ensure that subsequent risk voxels are formed around the actual site selection risk location, rather than being mechanically divided according to fixed dimensions, this embodiment first extracts boundary objects that will affect the artificial reef placement judgment from the registration dataset, then connects different boundaries as dividing lines, and finally uses the dividing lines to divide the candidate site selection area, so that the same risk voxel does not cross the location of sonar shadows, slope changes, seabed changes, or attitude disturbances; this can avoid a voxel containing both "landing areas" and "non-landing areas" at the same time, which would lead to mixed evidence when writing slots later; the implementation process includes the following steps.

[0039] In S2-1, after the registered dataset is input into the edge computing node, the edge computing node reads the registered sonar data, topographic data, seabed data, and pose data segment by segment along the track direction of the underwater survey vehicle. For sonar data, the edge computing node finds areas where continuous sonar echoes are interrupted or where echoes are significantly missing, and takes the boundary between the missing area and the effective echo area as the sonar shadow boundary. For topographic data, the edge computing node reads the seabed elevation changes along the track direction, and when the elevation changes from a gradual change to a zigzag or from a zigzag to a gradual change, it connects the turning points to form a slope toe line. For seabed data, the edge computing node reads adjacent positions... The substrate category is determined, and when adjacent locations switch from one category (hard bottom, sandy bottom, soft mud, etc.) to another, the category switch locations are connected to form a substrate change line. For pose data, edge computing nodes read the carrier's attitude change process within the track interval. When the attitude change direction within a certain track segment changes abruptly relative to the preceding and following track segments, that track segment is recorded as an attitude change segment. From this, the sonar shadow boundary, slope toe line, substrate change line, and attitude change segment are obtained, and a risk boundary set is output. Here, the attitude change segment is not used as separate site selection evidence, but as a segmentation basis to isolate data areas that may be affected by carrier sway.

[0040] In S2-2, after the risk boundary set is input into the edge computing node, the edge computing node first reads the positional relationships of various risk boundaries in the spatial plane of the candidate site selection area. When sonar shadow boundaries, slope toe lines, bottom sediment change lines, or attitude change segments intersect in the spatial projection, the edge computing node uses the intersection point as the connection point to connect the intersecting boundaries into the same dividing line. When two risk boundaries do not intersect but their ends fall into the same or adjacent spatial units to be divided, the edge computing node treats them as adjacent boundaries and connects them according to the shortest connection path between their ends. When the connected dividing lines are joined end to end, a closed dividing line is formed. When the connected dividing lines are not connected end to end but can separate a portion of the candidate site selection area, a semi-closed dividing line is formed. For a semi-closed dividing line, the edge computing node extends along its end extension direction or toward the nearest candidate site selection area boundary until it reaches the candidate site selection area boundary, avoiding the unclosed dividing line from remaining inside the voxel. This outputs a set of risk dividing lines. For example, if one end of the slope toe line is close to the sonar shadow boundary and the other end does not form a closed area, the end close to the sonar shadow boundary is connected to the shadow boundary, and the unclosed end is extended to the candidate site selection area boundary, so that the slope toe variation is not included in the same risk voxel.

[0041] In S2-3, after the risk segmentation line set is input into the edge computing node, the edge computing node uses the outer boundary of the candidate site selection area as the outermost segmentation boundary and superimposes the risk segmentation line set onto this outer boundary. Subsequently, the edge computing node spatially segments the candidate site selection area according to the position of the risk segmentation line. Any initial spatial block that is crossed by the sonar shadow boundary, slope toe line, bottom sediment change line, or attitude change segment continues to split along the corresponding risk segmentation line until each segmented spatial unit no longer crosses any of the above-mentioned risk boundaries. For areas not crossed by risk segmentation lines, the edge computing node retains its adjacent spatial range as a risk voxel. For areas where multiple risk segmentation lines intersect, the edge computing node generates multiple risk voxels with the intersection point as a common boundary, so that each risk voxel is located only on one side of the risk boundary. Finally, a risk voxel set is output. Each risk voxel in this risk voxel set retains its spatial boundary, adjacent voxel relationship, and which risk segmentation lines participated in the segmentation, for use when writing subsequent topographic trenches, bottom sediment trenches, flow trenches, and location trenches.

[0042] Through the above implementation process, the edge computing node does not divide the candidate site selection area according to a fixed grid size. Instead, it transforms sonar shadows, slope toe changes, bottom sediment changes, and attitude disturbances into voxel boundaries. This makes the data sources and site selection meanings within the same risk voxel more concentrated, reducing the possibility of conflicting evidence being mixed in the same voxel during subsequent trench writing. In practical applications: When an underwater survey vehicle passes through a gentle slope turning into a ditch area, sonar data shows echo loss on the back side of the ditch, topographic data shows an elevation change at the slope toe, and the bottom sediment identification result changes from hard bottom to soft mud. At the same time, the vehicle experiences a sudden attitude change during the turn. The edge computing node extracts the sonar shadow boundary, slope toe line, bottom sediment change line, and attitude change segment, connects these boundaries into dividing lines, and then divides the candidate site selection area with dividing lines. This ensures that hard bottom gentle slope, slope toe ditch, soft mud area, and attitude disturbance area fall into different risk voxels. During subsequent trench writing, evidence of hard bottom support and evidence of soft mud subsidence will not be written into the same voxel.

[0043] S3. For each risk voxel, establish a topographic trough, sediment trough, flow trough, and landing trough, and write the registration data falling into the risk voxel into the corresponding trough. The original measured data is written directly. When two data from different sources point to the same trough result, they are merged and written. Only the result obtained from adjacent risk voxels is downgraded and written. When data from different sources point to opposite trough results, a conflict marker is written. Output a set of risk voxels with trough markers.

[0044] To prevent the registration data from being simply averaged or smoothed after entering the risk voxel, this embodiment sets each risk voxel as an evidence unit with topographic, sedimentary, hydrological, and location troughs. The edge computing nodes first construct an observation matrix according to the trough, extract stable measured components within the same trough using singular value decomposition, and then complete the candidate writing results using regularized linear solution. Before writing, direct observations, merged observations, adjacent propagation, and opposite conflicts are distinguished to avoid the results generated by interpolation or smoothing being mistaken for true site selection evidence. This implementation process includes the following steps.

[0045] In S3-1, after the registration data falling within the same risk voxel is input into the edge computing node, the edge computing node first classifies it into topographic trenches, sedimentary trenches, hydrographic trenches, and location trenches according to the location meaning indicated by the data. The topographic trench receives data such as slope morphology, slope toe fractures, and support surface continuity; the sedimentary trench receives bottom surface identification data such as hard bottom, soft mud, and sandy bottom; the hydrographic trench receives data related to flow direction, bypass flow, direct flow, or backfilling; and the location trench receives data such as bottom surface contact, obstacle occupancy, and suspension risk. For each trench location, the edge computing node constructs an observation matrix using multiple data points from the same trench location within the same risk voxel. The rows of the matrix correspond to different acquisition times or different data sources, and the columns of the matrix correspond to the observation fields of that trench location; for example, the columns of the topographic trench can correspond to elevation. The columns of the substrate trench correspond to substrate category codes and surface texture responses based on changes in slope aspect and local support status. Subsequently, the edge computing node performs singular value decomposition on the observation matrix, taking the direction corresponding to the largest singular value as the main singular direction that most measured data in that trench point to, and retaining the components in the main singular direction as measured components. Components orthogonal to the main singular direction and prominent only in a few rows are removed as drift components, which usually come from carrier attitude jitter, local registration deviation, or single anomalous echoes. Through this processing, the edge computing node outputs the direct observation vector of each risk voxel in each trench. This direct observation vector still retains the original source information and is used to determine whether it belongs to direct writing or merged writing in subsequent processes.

[0046] In S3-2, after the direct observation vector is input into the edge computing node, the edge computing node establishes candidate write results to be solved for each risk voxel and each slot. During the solution process, the edge computing node uses the deviation between the candidate write results and the direct observation vector of that slot as a data item, uses the variation amplitude of the same slot in adjacent risk voxels as a regularization term, and records the residual source of the same data source separately as constraint information. This ensures that the solution result is close to the actual measured data of the voxel, but does not cover the existing observations of the voxel due to the smoothing effect of adjacent voxels. In specific execution, if a slot of a certain risk voxel has a direct observation vector, then the slot is solved primarily based on the direct observation vector. If a slot has no directly observed vector, but adjacent risk voxels have slot results with the same name, then the slot is only allowed to obtain candidate write results through the regularization term of adjacent risk voxels. If the data from the same source deviates continuously from the solved candidate write results, the edge computing node does not directly delete the source data, but records the residual direction and residual source corresponding to the source. Thus, the regularized linear solution algorithm outputs the candidate results and residual records for each risk voxel in the topographic slot, sediment slot, flow slot, and landing slot. Among them, the candidate results are used for subsequent write judgment, and the residual records are used for subsequent identification of cross-source consistency or cross-source conflict, rather than as a simple threshold screening basis.

[0047] In S3-3, after inputting candidate results and residual records into the edge computing node, the edge computing node compares the directions of candidate results from different sources in the same slot according to risk voxels and slot locations. Here, "direction" refers to the direction the slot results point towards the artificial reef site selection, not limited to geometric vector directions. For example, in a topographic slot, the direction is opposite to that of a continuous support surface and a slope toe fracture; in a sedimentary slot, the direction is opposite to that of a hard bottom and soft mud; in a flow channel, the direction is opposite to that of around-flow and direct flushing or backfilling; and in a placement slot, the direction is opposite to that of a complete bottom contact and a suspended bottom or obstacle-occupied location. If two candidate results from different sources point to the same slot result, for example, both sonar data and image data point to a hard bottom, or both sonar topography and pose-corrected geometric results point to a continuous support surface, then the edge computing node... The edge computing node merges the candidate results into the corresponding slot and retains the source markers of the merged results. If a slot does not have a direct observation vector of an ontology element, and its candidate results are only obtained by the regularization term of the adjacent risk voxels, the edge computing node downgrades the candidate results and marks them as the transmission results. If two candidate results from different sources point to opposite slot results, such as one source pointing to a hard bottom and the other to soft mud, or one source pointing to complete bottom contact and the other to obstacle occupation, the edge computing node does not average the two or cover one with the other, but writes a conflict marker in the corresponding slot. Through this process, the edge computing node outputs slot results with source markers, downgrade markers, and conflict markers.

[0048] In S3-4, when writing the slot results with source markers, degradation markers, and conflict markers back to the corresponding risk voxels, the edge computing node updates the risk voxels item by item according to the slot. For direct observation results and merged write results, the corresponding slot is written and the source marker is recorded. For degradation write results, the corresponding slot is written but the degradation marker is retained at the same time, so that the result can be used for 3D model display, missing certificate location, and supplementary survey task generation, but it cannot replace the direct observation results in the closure determination. For conflict markers, the edge computing node fixes the conflict marker in the corresponding slot and retains the source result that caused the conflict, and does not allow the smooth result obtained by subsequent regularized linear solution to cover the conflict marker. If the subsequent supplementary survey obtains new direct observations or cross-source consistent results, the edge computing node can update the corresponding slot according to the new direct observations or merged write results, but does not eliminate the conflict through regularization terms during this write process. Thus, a risk voxel set with slot markers is output. Each risk voxel in the risk voxel set includes topographic slots, sediment slots, flow slots, and landing slots, as well as the source marker, degradation marker, or conflict marker corresponding to each slot.

[0049] Through the above implementation process, the edge computing node does not directly fuse the data falling into the risk voxel into a single value. Instead, it first uses singular value decomposition to extract stable measured components within the voxel, then uses regularized linear solution to form candidate writing results, and finally writes merging, downgrading, and conflict markers according to three cases: consistent source, adjacent transmission, and opposite source. This can prevent missing data areas from being interpolated by adjacent data to form "seemingly complete" site selection evidence, and can also prevent opposite observations from losing risk indications after being averaged. In practical applications: a risk voxel is located in the candidate landing area of ​​an artificial reef. Near the toe of the slope, if sonar data shows a continuous supporting surface within the voxel and image data also shows a hard bottom, then the topographic trough and the bottom sediment trough can be merged and written. If there is no actual water flow measurement for this voxel, and the flow around the voxel is only obtained through the water flow trough of the adjacent voxel, then the water flow trough can only be downgraded and cannot prove the closure of this voxel on its own. If sonar data from another segment shows that the same voxel has a slope toe fracture, which is contrary to the continuous supporting surface result of the previous segment, then the topographic trough is written with a conflict mark. Subsequent closure determination will not treat this voxel as a closed voxel, but will retain the conflict until the supplementary measurement task generation stage.

[0050] S4. Perform closure determination on risk voxel sets with slot markings: When there are direct write or merged write results in the topographic slot, bottom sediment slot, water flow slot and landing slot, and there are no conflict markings in the four slots, the corresponding risk voxel is recorded as a closed voxel; otherwise, it is recorded as an unclosed voxel, and the missing certificate slot is recorded. The closed voxel, unclosed voxel and missing certificate slot are output.

[0051] To enable the risk voxel set with slot markings to be transformed into a closed result that can be used for site selection judgment, this embodiment does not use the continuity of three-dimensional geometry as the criterion for judgment. Instead, it checks one by one whether the topographic slot, bottom sediment slot, flow slot, and location slot in each risk voxel have sufficient evidence to support the site selection. Among them, direct writing results and merged writing results are from voxel measurements or cross-source consistent observations and are considered valid evidence. Degraded writing results, empty slots, and conflict markers cannot support the site selection on their own and are considered non-closed evidence. Subsequently, the edge computing nodes record the missing slots in a fixed slot order and merge spatially adjacent unclosed voxels with the same missing information type to facilitate the generation of subsequent supplementary measurement tasks. This implementation process includes the following steps.

[0052] In S4-1, after the risk voxel set with slot markings is input into the edge computing node, the edge computing node reads the topographic slot, substrate slot, flow slot, and landing slot for each risk voxel one by one, and checks the write type in each slot. When the slot contains a direct write result, it indicates that the slot contains original measured data falling into this risk voxel, such as slope results directly generated by sonar or flow direction results directly generated by flow sensors, and the edge computing node records it as valid evidence. When the slot contains a merged write result, it indicates that the slot contains data from two or more different sources that point to the same slot result, such as both sonar and image pointing to a hard bottom, and the edge computing node also records it as valid evidence. When a slot contains a downgraded write result, it indicates that the result was only obtained from an adjacent risk voxel and cannot prove that the risk voxel itself possesses the evidence. The edge computing node records this as non-closed evidence. When a slot is empty, it indicates that there is no usable write result in the slot. The edge computing node records this as non-closed evidence. When a conflict marker exists in a slot, it indicates that different sources have given opposite addressing directions to the slot. The edge computing node records this as non-closed evidence and retains the conflict source. Thus, the edge computing node generates a slot evidence table for each risk voxel. The slot evidence table records at least the voxel number, slot name, slot result, write type, evidence status, and source marker.

[0053] In S4-2, after the slot evidence table is input into the edge computing node, the edge computing node performs closure verification on each risk voxel in the order of topographic slot, sediment slot, flow slot, and landing slot. This fixed order is used to generate subsequent missing evidence slot records and does not indicate that one slot can replace another. The edge computing node first reads the evidence status of the topographic slot. If it is valid evidence, it continues to read the sediment slot; if it is non-closed evidence, it writes the topographic slot into the missing evidence slot of that risk voxel. Then, it reads the sediment slot, flow slot, and landing slot in sequence and records each non-closed evidence slot. When one If all four slots of a risk voxel are valid evidence and there are no conflict markers in any of the four slots, the edge computing node records the risk voxel as a closed voxel. If any slot is a downgraded write, an empty slot, or a conflict marker, the edge computing node records the risk voxel as an unclosed voxel and records the corresponding slot as a missing evidence slot. If the same risk voxel has multiple unclosed evidence slots, multiple missing evidence slots are recorded in the above fixed order. The initial closure determination result is then output, which includes the closure status of each risk voxel and the missing evidence slots corresponding to the unclosed voxels.

[0054] In S4-3, after the initial closure determination result is input into the edge computing node, the edge computing node reads the adjacency relationship between risk voxels and performs connectivity merging on the missing slots of unclosed voxels. The adjacency relationship comes from the previous risk voxel segmentation result. Risk voxels with a common face, common edge, or common angle can be identified as spatially adjacent voxels. When used for supplementary path generation, voxels with adjacent common faces are preferentially merged. The edge computing node first selects an unclosed voxel as the starting voxel, reads its missing slots, then searches for other unclosed voxels spatially adjacent to the starting voxel and with the same missing slots, and connects the searched unclosed voxels... The merged voxels continue to expand outwards as a new starting point until no new adjacent unclosed voxels with the same missing slots are found. Subsequently, the edge computing node merges these unclosed voxels into the same missing slot block, while retaining the missing slot record for each unclosed voxel itself. If adjacent unclosed voxels have different missing slots, such as one lacking a substrate slot and the other lacking a flow slot, they are not merged into the same missing slot block. If an unclosed voxel lacks multiple slots simultaneously, it participates in the merging record of the corresponding missing slots separately. Thus, the edge computing node outputs closed voxels, unclosed voxels, and missing slots, and can also include the missing slot block as the organization object for subsequent supplementary testing tasks.

[0055] Through the above implementation process, edge computing nodes can transform risk voxel sets marked with slot locations into closed voxels and unclosed voxels, and clearly identify whether each unclosed voxel lacks a topographic slot, sediment slot, flow slot, or location slot. Simultaneously, unclosed voxels that are spatially adjacent and have the same missing slot location are grouped into a missing-slot block, facilitating subsequent centralized supplementary measurements according to the missing slot type, rather than repeatedly supplementing all unclosed voxels one by one. In practical applications: if multiple adjacent risk voxels in a slope toe area within a candidate site have direct write results for both topographic and location slots, but the sediment slot is empty, these risk voxels will be recorded as unclosed voxels and grouped into the same sediment missing-slot block. If another adjacent risk voxel, although also located near the slope toe, has a conflict marker in its flow slot, this voxel will not be merged into the sediment missing-slot block, but will instead enter the subsequent supplementary measurement process as a flow missing-slot or conflict object, thus enabling the supplementary measurement task to distinguish between surface re-sampling and flow direction re-measurement.

[0056] The core logic of the aforementioned closure determination is to verify the four core criteria required for site selection decisions as independent dimensions. Only when all four criteria have valid evidence from actual measurements or cross-source consistency is the site selection evidence for that area considered closed. This is fundamentally different from the traditional modeling logic of "complete if there are no voids on the geometric surface": traditional methods focus on the visual integrity of the model, while this approach focuses on the sufficiency of evidence for decision-making. Even if the geometric surface of the model is continuous, if any type of site selection criterion is missing evidence or there is a conflict, the area will still be marked as unclosed, thus avoiding misjudgment that substitutes visual integrity for decision reliability.

[0057] S5. Generate a 3D model based on closed voxels, unclosed voxels, and missing evidence slots. Then, perform a site selection judgment for each unclosed voxel after replacing it with the favorable and unfavorable results of the missing evidence slot. If the two judgment results are different, list the unclosed voxel as a supplementary test object and output the evidence closure type 3D model and the supplementary test task.

[0058] To transform the closure determination results into a three-dimensional model applicable to artificial reef site selection, this embodiment first organizes closed and unclosed voxels into voxel maps, then generates three-dimensional model units from the voxel maps, and writes the slot status of each voxel into the corresponding model unit; subsequently, favorable and unfavorable results are written into the missing slots of unclosed voxels to form two replacement models, and the impact of the two replacement models on the site selection conclusion of the same candidate site is compared. Only unclosed voxels that would change the site selection conclusion are listed as supplementary measurement objects; the purpose of this process is to avoid blindly supplementing measurements in all missing areas, but to prioritize supplementing measurements in key missing locations that would affect whether artificial reefs can be placed; this implementation process includes the following steps.

[0059] In S5-1, after inputting closed voxels, open voxels, and missing slots into the edge computing node, the edge computing node treats each voxel as a node in the voxel graph and reads the spatial boundaries of each voxel. When two voxels have a common surface, the edge computing node establishes a connecting edge between the corresponding nodes of the two voxels, indicating that they are coplanarly adjacent in three-dimensional space. Subsequently, the edge computing node reads the topographic slot, substrate slot, water flow slot, and location slot for each voxel and converts each slot into a slot state. The slot state includes at least... This includes the slot name, slot result, writing type, presence of conflict markers, and whether it is a slot with missing certification. For example, if a closed voxel's topographic slot is a directly written continuous support surface, the sediment slot is a merged written hard bottom, the water flow slot is a directly written flow around, and the landing slot is a directly written bottom surface with complete contact, then this voxel is recorded as a closed node in the voxel graph. If a non-closed voxel's sediment slot is an empty slot, then this voxel is recorded as a non-closed node with missing sediment slot certification in the voxel graph. This outputs the slot status diagram.

[0060] In S5-2, after the slot status map is input into the edge computing node, the edge computing node generates 3D model units according to the spatial boundaries of each voxel, with each voxel corresponding to one 3D model unit. For two voxels connected by coplanar connecting edges in the voxel map, the edge computing node treats the common face of their corresponding 3D model units as an internal connecting face when generating the 3D model, so that the 3D model can geometrically continuously express the spatial morphology of the candidate site selection area. It should be noted that the merging of the common face is only used for geometric expression and does not change the slot status of each of the two 3D model units. That is, if one 3D model unit is in a closed state and the other 3D model unit has a missing slot, even if they are coplanar adjacent, the closed state or slot results will not overlap. Subsequently, the edge computing node writes the slot status of the topographic slot, bottom sediment slot, water flow slot and landing slot of each voxel into the corresponding 3D model unit, so that each 3D model unit has both geometric position and evidence status, and outputs an evidence-closed 3D model.

[0061] In S5-3, after the evidence-closed 3D model is input into the edge computing node, the edge computing node reads the 3D model elements containing unclosed voxels one by one, and copies the local model with the candidate placement area of ​​the unclosed voxel as the range. During copying, the slot status of adjacent closed voxels in the local model remains unchanged, and only the missing evidence slots of the current unclosed voxel are allowed to be replaced. For the same unclosed voxel, the edge computing node generates favorable replacement models and unfavorable replacement models respectively: when the terrain trench lacks evidence, the favorable result is written as a continuous support surface, and the unfavorable result is written as a slope toe fracture; when the bottom sediment trench lacks evidence, the favorable result is written as... For a hard bottom, the unfavorable result is written as soft mud; when the flow channel lacks certification, the favorable result is written as bypass flow, and the unfavorable result is written as direct flow or backfilling; when the placement channel lacks certification, the favorable result is written as complete bottom surface contact, and the unfavorable result is written as bottom surface suspended or obstructed; if an unclosed voxel has multiple uncertified slots, the edge computing nodes perform the above replacement on each uncertified slot and retain the slot name corresponding to the replacement; the favorable and unfavorable results here are used to test whether the uncertified slot affects the site selection conclusion, and do not indicate that the true state of the uncertified slot has been determined; thus, favorable replacement model and unfavorable replacement model are output.

[0062] In S5-4, after the favorable and unfavorable replacement models are input into the edge computing node, the edge computing node determines the bottom surface occupancy range of the artificial reef based on the bottom contour, predetermined posture, and candidate landing area location of the artificial reef to be deployed, and reads the 3D model elements covered or contacted by the bottom surface occupancy range. For the favorable replacement model, the edge computing node checks whether the slots involved in the site selection in the read 3D model elements meet the landing requirements, i.e., the topographic slot should be a continuous support surface, the bottom substrate slot should be a hard bottom, the water flow slot should be a bypass channel, and the landing slot should be a bottom... The surface contact is complete and there are no conflict markers. If all three-dimensional model units within the bottom surface area meet the above conditions, the candidate location area is judged as selectable. For unfavorable replacement models, the edge computing nodes read three-dimensional model units at the same location within the same bottom surface area. If any three-dimensional model unit has slope toe fracture, soft mud, direct impact, backfilling, bottom surface suspension, obstacle occupation, missing certificate slot, or conflict markers, the candidate location area is judged as unselectable. Thus, the site selection judgment results corresponding to favorable replacement models and unfavorable replacement models are obtained.

[0063] In S5-5, after inputting the two site selection judgment results into the edge computing node, the edge computing node compares the judgment results of the same candidate site area under the favorable replacement model and the unfavorable replacement model. If the favorable replacement model determines that the site is selectable, while the unfavorable replacement model determines that it is not selectable, it means that the missing certificate slot of the unclosed voxel will change the site selection conclusion of the candidate site area, and the edge computing node lists the unclosed voxel as a supplementary measurement object. If both judgment results are selectable or both are not selectable, it means that the current missing certificate slot of the unclosed voxel will not change the site selection conclusion of the candidate site area. The site selection conclusion is that edge computing nodes retain their missing data slots but do not prioritize them for supplementary surveys. For unclosed voxels listed as supplementary survey targets, edge computing nodes generate supplementary survey tasks according to the missing data slots: sonar rescanning tasks are generated when topographic voxels lack data, bottom surface resampling tasks are generated when sediment voxels lack data, flow direction rescanning tasks are generated when flow channels lack data, and surround rescanning tasks are generated when location voxels lack data. Each supplementary survey task includes at least the location of the supplementary survey target, the missing data slot, the data type of the supplementary survey, and the supplementary survey path type. This outputs a closed-loop 3D model of evidence and the supplementary survey task.

[0064] Through the above implementation process, the edge computing node transforms closed and open voxels into a 3D model with slot status. It then uses favorable and unfavorable replacements to determine whether missing slots will overturn the site selection conclusion for the same candidate site, thus limiting the supplementary measurement objects to open voxels that truly affect the artificial reef site selection, reducing invalid supplementary measurements. In practical applications: if an open voxel exists within a candidate site, and its topographic and site slots already have direct writing results, but the bottom sediment slot lacks certification, the edge computing node first writes the bottom sediment slot into the hard bottom in the favorable replacement model and performs the selection process. If all other model units within the bottom surface area meet the placement requirements, the site is deemed a candidate. Subsequently, the bottom sediment trench is written into the soft mud in the unfavorable replacement model and re-evaluated. If the candidate site is deemed unselectable due to the risk of soft mud subsidence, it indicates that the lack of evidence for the bottom sediment trench will change the site selection conclusion. The edge computing node lists the unclosed voxel as a supplementary measurement object and generates a bottom surface re-sampling task for that location. Conversely, if both determinations are unselectable, for example, if there is already a slope toe fracture or obstacle in the area, the lack of evidence for the bottom sediment will not change the current site selection conclusion and can be temporarily not considered as a priority supplementary measurement object.

[0065] Furthermore, it also includes a 3D modeling system for underwater surveying and site selection of artificial reefs, including a registration and assembly module, a segmentation module, a trench marking module, a closure and missing identification module, and a modeling and supplementary survey module.

[0066] The registration and set module is used to acquire multi-source survey data of candidate site selection areas, and the edge computing nodes align each data to the same coordinate system according to the track time scale, and output the registration dataset.

[0067] The voxel segmentation module is used to input the registration dataset into the edge computing node. The edge computing node extracts the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment. Using the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment as voxel boundaries, the candidate site selection area is segmented and the risk voxel set is output.

[0068] The slot labeling module is used to create topographic slots, sediment slots, flow slots, and landing slots for each risk voxel, and write the registration data falling into the risk voxel into the corresponding slot. The original measured data is written directly, when two data from different sources point to the same slot result, they are merged and written, only the result obtained from adjacent risk voxels is downgraded and written, and when data from different sources point to opposite slot results, a conflict marker is written, and a set of risk voxels with slot markers is output.

[0069] The closure and deficiency identification module is used to perform closure determination on risk voxel sets with slot markings: when there are direct write or merged write results in the topographic slot, bottom sediment slot, water flow slot and landing slot, and there are no conflict markings in the four slots, the corresponding risk voxel is recorded as a closed voxel; otherwise, it is recorded as an unclosed voxel, and the missing certificate slot is recorded, and the closed voxel, unclosed voxel and missing certificate slot are output.

[0070] The modeling and supplementary testing module is used to generate a 3D model based on closed voxels, unclosed voxels, and missing evidence slots. For each unclosed voxel, it performs a site selection judgment after replacing it with the favorable and unfavorable results of the missing evidence slot. If the two judgment results are different, the unclosed voxel is listed as a supplementary testing object, and the evidence closure type 3D model and supplementary testing task are output.

[0071] Working principle: The underwater survey vehicle first collects data from candidate site selection areas, and the edge computing nodes then align the data to the same coordinate system according to the flight time. Subsequently, the system identifies locations that are not visible to sonar, locations of slope changes, locations of bottom changes, and locations of abrupt attitude changes, and uses these locations as boundaries to segment risk voxels. Each risk voxel records topography, bottom sediment, current, and landing information, and writes evidence according to its source. Measured results are written directly, results from multiple sources that are consistent are merged, results derived from adjacent voxels are downgraded, and results that are opposite are marked with a conflict flag. After that, the system determines whether the four types of information for each risk voxel are complete and conflict-free. If they are satisfied, it is recorded as a closed voxel; otherwise, it is recorded as an open voxel and missing items are recorded. Finally, the system generates a 3D model and fills in the missing items of open voxels with favorable and unfavorable results to perform two site selection judgments. If the two judgments would change the deployment conclusion, a corresponding supplementary survey task is generated.

[0072] For example, when selecting a site in turbid waters, if sonar cannot clearly see behind a certain slope, and the image shows that the bottom surface may change from hard to soft mud, the system will segment that location into risk voxels and find that there is evidence of its topography and placement information, but the bottom sediment information is missing. At this time, the system will not directly determine that the location is suitable for deployment, but will judge it separately according to hard bottom and soft mud. If it is suitable for deployment in hard bottom but not in soft mud, it means that the missing bottom sediment will affect the conclusion, and the system will generate a bottom surface re-sampling task. If it is not suitable for deployment in either hard bottom or soft mud, the missing item will not be prioritized for re-measurement. In this way, the 3D model not only shows the seabed morphology, but also points out which locations have insufficient evidence and which locations must be re-measured.

[0073] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A three-dimensional modeling method for underwater surveying and site selection of artificial reefs, characterized in that, include: S1. Obtain multi-source survey data of the candidate site selection area, and have the edge computing nodes align each data to the same coordinate system according to the track time scale, and output the registration dataset; S2. Input the registration dataset into the edge computing node. The edge computing node extracts the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment. Using the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment as voxel boundaries, the candidate site selection area is segmented and the risk voxel set is output. S3. Create a topographic trough, substrate trough, water flow trough, and landing trough for each risk voxel, and write the registration data that falls into the risk voxel into the corresponding trough. The original measured data is written directly; when two data sources from different sources point to the same slot result, they are merged and written; the result obtained only by adjacent risk voxels is downgraded and written; when data from different sources point to opposite slot results, a conflict marker is written; and a risk voxel set with slot markers is output. S4. Perform closure determination on the risk voxel set with slot markings: When there are direct write or merged write results in the topographic slot, bottom sediment slot, flow slot and landing slot, and there are no conflict markings in the four slots, the corresponding risk voxel is recorded as a closed voxel; otherwise, it is recorded as an unclosed voxel, and the missing certificate slot is recorded. The closed voxel, unclosed voxel and missing certificate slot are output. S5. Generate a 3D model based on closed voxels, unclosed voxels, and missing evidence slots. Then, perform a site selection judgment for each unclosed voxel after replacing it with the favorable and unfavorable results of the missing evidence slot. If the two judgment results are different, list the unclosed voxel as a supplementary test object and output the evidence closure type 3D model and the supplementary test task.

2. The three-dimensional modeling method for underwater surveying and site selection of artificial reefs according to claim 1, characterized in that, S1 includes: S1-1. Input the multi-source survey data of the candidate site selection area into the edge computing node. The edge computing node uses the track time of the underwater survey carrier as the main sequence, writes the acquisition time of each survey data between the two adjacent track points, and outputs the survey dataset with track interval markings. S1-2. Input the survey dataset with track interval markers into the edge computing node. The edge computing node calculates the registration pose of the survey data according to the time position of each survey data in the corresponding track interval, along the pose change direction of the track points at both ends of the track interval, and projects the survey data to the same coordinate system according to the registration pose, and outputs the initial registration dataset. S1-3. Input the initial registration dataset into the edge computing node. The edge computing node retains the original source, track interval label and registration pose of the survey data projected to the same spatial location but with different track interval labels. It does not merge the survey data into a single result and outputs the registration dataset for subsequent risk voxel writing.

3. The three-dimensional modeling method for underwater surveying and site selection of artificial reefs according to claim 2, characterized in that, S2 includes: S2-1. Input the registration dataset into the edge computing node. The edge computing node reads the sonar echo missing area, seabed elevation turning area, seabed sediment type switching area and pose change area along the track direction of the underwater survey vehicle. It then extracts the corresponding sonar shadow boundary, slope line, seabed sediment change line and pose change segment, and outputs the risk boundary set. S2-2. Input the risk boundary set into the edge computing node. The edge computing node connects the intersecting or adjacent sonar shadow boundaries, slope toe lines, bottom change lines and attitude change segments into closed or semi-closed dividing lines. The unclosed dividing lines are extended to the boundary of the candidate site selection area, and the risk dividing line set is output. S2-3. Input the risk segmentation line set into the edge computing node. The edge computing node uses the risk segmentation line set as the voxel segmentation boundary to perform non-uniform spatial segmentation of the candidate site selection area, so that the same risk voxel does not cross the sonar shadow boundary, slope toe line, bottom change line or attitude change segment, and outputs the risk voxel set.

4. The three-dimensional modeling method for underwater surveying and site selection of artificial reefs according to claim 3, characterized in that, S3 includes: S3-1. Input the registration data falling into the same risk voxel into the edge computing node. The edge computing node constructs the observation matrix according to the topographic trough, bottom sediment trough, flow trough and location trough respectively. Then, it performs singular value decomposition on the observation matrix to retain the measured components corresponding to the main singular direction and remove the drift components orthogonal to the main singular direction. The direct observation vector of each trough is output. S3-2. Input the direct observation vector into the edge computing node. The edge computing node uses a regularized linear solution algorithm to solve the candidate writing results of each slot, constrained by the consistency of slot results, the continuous change of adjacent risk voxels, and the limitation of residuals from the same source. It outputs the candidate results and residual records of each risk voxel in the topographic slot, bottom sediment slot, water flow slot and landing slot.

5. The three-dimensional modeling method for underwater surveying and site selection of artificial reefs according to claim 4, characterized in that, S3 also includes: S3-3. Input the candidate results and residual records into the edge computing node. The edge computing node compares the directions of candidate results from different sources in the same slot. If the directions are consistent, they are merged and written while retaining the source mark. If the candidate result is only obtained by the regularization term of the adjacent risk voxel, it is downgraded and written as a transmission result. If the directions are opposite, merging is not performed and a conflict mark is written. The slot result with source mark, downgrade mark and conflict mark is output. S3-4. Write the slot results with source markers, degradation markers and conflict markers back to the corresponding risk voxels. The edge computing nodes prohibit degradation markers from replacing direct observation results alone and prohibit conflict markers from being covered by regularization results. Output the risk voxel set with slot markers.

6. The three-dimensional modeling method for underwater surveying and site selection of artificial reefs according to claim 5, characterized in that, S4 includes: S4-1. Input the risk voxel set with slot markings into the edge computing node. The edge computing node reads the topographic slot, bottom sediment slot, water flow slot and landing slot of each risk voxel one by one. The direct write results and merged write results in each slot are recorded as valid evidence. The downgraded write results, empty slots and conflict markers are recorded as non-closed evidence. The slot evidence table of each risk voxel is output.

7. The three-dimensional modeling method for underwater surveying and site selection of artificial reefs according to claim 6, characterized in that, S4 also includes: S4-2. Input the slot evidence table into the edge computing node. The edge computing node performs closure verification in a fixed order of topographic slot, bottom sediment slot, water flow slot and landing slot. If all four slots are valid evidence and there are no conflict markers, the corresponding risk voxel is recorded as a closed voxel. If any slot is non-closed evidence, the corresponding risk voxel is recorded as an unclosed voxel and the slot is extracted as a missing evidence slot. The initial closure judgment result is output. S4-3. Input the initial closure judgment result into the edge computing node. The edge computing node connects and merges the missing certificate slots of adjacent risk voxels, and merges unclosed voxels with the same missing certificate slots and spatially adjacent voxels into the same missing certificate block. Output closed voxels, unclosed voxels and missing certificate slots.

8. The three-dimensional modeling method for underwater surveying and site selection of artificial reefs according to claim 7, characterized in that, S5 includes: S5-1. Input closed voxels, unclosed voxels, and missing voxels into the edge computing node. The edge computing node will build a voxel map with each voxel as a node and coplanar adjacent voxels as connecting edges. The topographic voxel, bottom sediment voxel, water flow voxel, and landing voxel of each voxel will be converted into the voxel position state and the voxel position state map will be output. S5-2. Input the slot status map into the edge computing node. The edge computing node generates three-dimensional model elements according to the voxel space boundary. The coplanar adjacent three-dimensional model elements are merged into their common surface. The slot status of the corresponding voxel is written into the corresponding three-dimensional model element. The evidence closed three-dimensional model is output. S5-3. Input the evidence-closed 3D model into the edge computing node, and have the edge computing node copy the local model containing unclosed voxels one by one; write favorable and unfavorable results for the missing evidence slots of the same unclosed voxel respectively. The favorable result for the topographic slot is a continuous support surface, and the unfavorable result is a slope toe fracture. The favorable result for the bottom slot is a hard bottom, and the unfavorable result is soft mud. The favorable result for the water flow slot is bypass flow, and the unfavorable result is direct flow or back siltation. The favorable result for the placement slot is complete bottom surface contact, and the unfavorable result is bottom surface suspension or obstacle occupation. Output favorable and unfavorable replacement models.

9. The three-dimensional modeling method for underwater surveying and site selection of artificial reefs according to claim 8, characterized in that, S5 also includes: S5-4. Input the favorable replacement model and unfavorable replacement model into the edge computing node. The edge computing node reads the three-dimensional model units in the same candidate placement area according to the occupation range of the bottom surface of the artificial reef. When all the read three-dimensional model units have favorable results and no conflict markers, they are judged as selectable. When any three-dimensional model unit has unfavorable results, missing slots, or conflict markers, it is judged as unselectable. Output the two site selection judgment results. S5-5. Input the results of the two site selection judgments into the edge computing node. When the edge computing node determines that one candidate site is selectable and the other is not selectable in the same candidate site area, list the corresponding unclosed voxels as supplementary measurement objects and generate supplementary measurement tasks according to their missing evidence slots. Among them, the topographic slot corresponds to sonar rescanning, the sediment slot corresponds to bottom surface resampling, the water flow slot corresponds to flow direction rescanning, and the site slot corresponds to surround rescanning. Output the evidence closure type 3D model and supplementary measurement tasks.

10. A three-dimensional modeling system for underwater surveying and site selection of artificial reefs, comprising a registration and assembly module, a segmentation module, a trench marking module, a closure and missing information identification module, and a modeling and supplementary survey module, characterized in that, The registration and set-up module is used to acquire multi-source survey data of candidate site selection areas, and the edge computing nodes align each data to the same coordinate system according to the track time scale, and output the registration dataset; The voxel segmentation module is used to input the registration dataset into the edge computing node. The edge computing node extracts the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment. Using the sonar shadow boundary, slope toe line, bottom sediment change line and attitude change segment as voxel boundaries, the candidate site selection area is segmented and the risk voxel set is output. The registration module is used to create topographic, sediment, flow, and placement slots for each risk voxel, and write the registration data that falls into the risk voxel into the corresponding slot. The original measured data is written directly; when two data sources from different sources point to the same slot result, they are merged and written; the result obtained only by adjacent risk voxels is downgraded and written; when data from different sources point to opposite slot results, a conflict marker is written; and a risk voxel set with slot markers is output. The closure and deficiency identification module is used to perform closure determination on risk voxel sets with slot markings: when there are direct write or merged write results in the topographic slot, bottom sediment slot, water flow slot and landing slot, and there are no conflict markings in the four slots, the corresponding risk voxel is recorded as a closed voxel; otherwise, it is recorded as an unclosed voxel, and the missing certificate slot is recorded, and the closed voxel, unclosed voxel and missing certificate slot are output. The modeling and supplementary testing module is used to generate a 3D model based on closed voxels, unclosed voxels, and missing evidence slots. For each unclosed voxel, it performs a site selection judgment after replacing it with the favorable and unfavorable results of the missing evidence slot. If the two judgment results are different, the unclosed voxel is listed as a supplementary testing object, and the evidence closure type 3D model and supplementary testing task are output.

Images