Deep sea LIBS rare earth element quantitative detection method and system fused with spatial information

Abstract

The invention provides a deep sea LIBS rare earth element quantitative detection method and system fused with spatial information, and belongs to the technical field of element analysis and detection based on deep learning. The method comprises the following steps: firstly, designing an integrated deep-sea LIBS sampling device, and realizing in-situ sampling of deep-sea sediments through an ROV; then, a multi-point sampling experiment is carried out according to a specific deep sea area sampling strategy, and a rare earth detection data set containing spectral characteristics, spatial position information and concentration labels is obtained; then, taking the sampling points as graph structure nodes, and fusing the spectral data and the spatial information to construct spectrum-bit fused graph structure data; a rare earth element quantitative detection model is constructed, an information propagation mechanism in a graph structure is utilized to capture an internal rule of rare earth elements in spatial distribution, and a region polymerization method is combined to realize recognition of a rare earth enrichment region; according to the invention, in-situ detection of rare earth elements and real-time identification of an enrichment area in a deep sea environment are realized, and intelligent technical support is provided for deep sea resource detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of element analysis and detection based on deep learning, and particularly relates to a method and system for quantitative detection of rare earth elements in deep sea LIBS by integrating spatial information. Background Technique

[0002] The complex deep-sea environment, high pressure, low visibility, and intense water body disturbance pose great challenges to the in-situ detection and quantitative analysis of rare earth elements. Most of the current underwater element detection methods face problems such as poor in-situ performance, strong signal interference, low detection accuracy, and lack of spatial information, making it difficult to meet the requirements of efficient, accurate, and automated exploration of rare earth resources. The existing methods for quantitative detection of rare earth elements mainly include the following three types:

[0003] Laboratory post-processing analysis method: This method combines inductively coupled plasma mass spectrometry (ICP-MS) for quantitative detection of rare earth elements and is one of the most widely used techniques in current rare earth element detection. After collecting deep-sea sediment samples, complex pre-treatment processes such as acid dissolution and purification are carried out in the laboratory, and then ICP-MS is used for highly sensitive quantitative analysis. Although its detection accuracy is high, this method highly depends on sample recovery and laboratory conditions and cannot achieve in-situ real-time detection. At the same time, the sample may be contaminated or its composition may change during transportation and processing, affecting the representativeness and reliability of the data. In addition, this method has high costs and long time consumption and is not suitable for large-scale and rapid underwater resource investigation tasks.

[0004] Underwater laser-induced breakdown spectroscopy (LIBS) in-situ detection method: Due to its in-situ detection ability, this method has been attempted to be applied to underwater element identification and sediment characterization. Related technologies are mostly based on integrating the LIBS probe into a submersible platform to achieve laser excitation and spectral acquisition of the sediment surface. However, traditional underwater LIBS devices mainly directly excite the sediment surface and lack the ability to sample and enrich internal components, resulting in the rare earth element signal being easily affected by environmental factors such as sediment coverage and seawater disturbance, with low detection signal-to-noise ratio and poor stability. In addition, most existing underwater LIBS systems are single-point detections and do not integrate spatial information, making it difficult to achieve regional-level element distribution modeling and enrichment trend identification.

[0005] Machine learning-based single-point modeling method: This method attempts to apply machine learning methods to the prediction of rare earth element concentrations in LIBS spectral data, and uses algorithms such as support vector regression (SVR) and random forest to model the collected single-point spectral data. To a certain extent, such methods have improved the intelligent level of data processing. However, most of their models are constructed based on single-point spectral features, fail to consider the spatial correlation between sampling points, and lack the ability to express regional structural information. At the same time, such models are easily affected by the limitations of training samples, have insufficient generalization ability, are difficult to maintain prediction stability and accuracy in complex deep-sea geological backgrounds, and have not yet formed an integrated detection system that can be deployed and operated in deep-sea environments.

[0006] Therefore, existing deep-sea rare earth element detection technologies have varying degrees of limitations in terms of in-situ detection, accuracy, anti-interference ability, and spatial information integration, and are difficult to meet the current requirements for efficient, accurate, and automated deep-sea resource exploration. In this context, there is an urgent need to develop an integrated technical path with in-situ detection capabilities, high-sensitivity analysis performance, and regional modeling and expression capabilities to break through the limitations of traditional methods and achieve rapid, stable, and intelligent identification of deep-sea rare earth elements, thereby providing strong technical support for the scientific assessment and strategic development of seabed resources. Summary of the Invention

[0007] In view of the above problems, the present invention proposes an integrated technical path with in-situ detection capabilities, high-sensitivity analysis performance, and regional modeling and expression capabilities to break through the limitations of traditional methods and achieve rapid, stable, and intelligent identification of deep-sea rare earth elements, thereby providing strong technical support for the scientific assessment and strategic development of seabed resources.

[0008] The first aspect of the present invention provides a method for quantitatively detecting rare earth elements in deep-sea LIBS by integrating spatial information, including the following processes:

[0009] S1, based on an integrated deep-sea LIBS sampling device and a regional sampling path and point distribution strategy, perform multi-point deep-sea sediment sampling in the target area to obtain rare earth element samples to be measured in the target area, record the corresponding spatial position information, and then detect the spectral characteristics of each sample to be measured by laser excitation;

[0010] S2, use different sampling points as nodes Node in the graph structure, combine the spectral characteristics and spatial distances of the sampling points to construct the edge Edge structure of the graph, encode the spectral characteristics and spatial position information of each sampling point as node features, and convert the sample data to be measured into graph structure data as the standard input data for the rare earth element concentration detection model;

[0011] S3. Input the graph structure data in S2 into the rare earth element concentration detection model based on the graph neural network (GNN) that has been constructed and trained. The model extracts the spatial correlation features of rare earth elements through the node feature propagation and aggregation mechanism of the GNN, and outputs the rare earth element concentration at each sampling point.

[0012] S4. Based on the rare earth element concentration at each sampling point within the target area, use the regional aggregation algorithm to calculate the regional-level rare earth enrichment trend and output the enrichment level.

[0013] Preferably, the specific structure and sampling process of the integrated deep-sea LIBS sampling device are as follows:

[0014] The device includes a sample chamber, a reaction chamber, an ROV manipulator, four deep-water motors and their corresponding piston structures.

[0015] Among them, the fourth deep-water motor is connected to the sampling piston through a slider and is controlled by the manipulator; the third deep-water motor is connected to the acid piston through a slider. The right chamber of the acid piston is connected to the reaction chamber through a pipeline, and an acid addition port is provided in the connecting chamber; a stirring and drainage integrated pump assembly is provided at the bottom of the reaction chamber; the reaction chamber is connected to the sample chamber through a pipeline and a stop valve; among them, the first deep-water motor is connected to the suction filtration and decompression piston through a slider. The right chamber of the suction filtration and decompression piston is connected to the sample chamber through a pipeline, and a pressure sensor is provided on the sample chamber; the second deep-water motor is connected to the seawater piston through a slider. The right chamber of the seawater piston is connected to the left chamber of the suction filtration and decompression piston through a pipeline; a seawater inlet is provided in the left chamber of the seawater piston.

[0016] Through the operation of the ROV manipulator, insert the sampling piston driven by the fourth deep-water motor into the deep-sea sediment in the predetermined area, and the motor drives the piston to draw backward to complete the collection of the sediment sample; then insert the piston into the reaction chamber and drive it in the reverse direction to push the sediment into the reaction chamber.

[0017] The third deep-water motor drives the acid piston to inject a preset volume of quantitative acid into the reaction chamber for sample pretreatment.

[0018] Start the stirring and drainage integrated pump to drive the impeller in the chamber to rotate, so that the acid and the sediment are fully mixed and reacted, and the rare earth elements in the sediment are fully dissolved into the acid to achieve component release.

[0019] After the mixing reaction is completed, the first deep-water motor drives the piston to extract the processed mixed sample in the reaction chamber into the special sample chamber for LIBS detection.

[0020] After closing the electric shut-off valve in the sample chamber, the first deep-water motor continues to pump. Due to the sealing of the system, the pressure in the chamber will gradually decrease. The pressure in the sample chamber is monitored by a pressure sensor. When the pressure in the sample chamber drops to the set value, the sample is at the target pressure and in a sealed state. As a test sample for LIBS detection, it can be directly used in laser excitation experiments.

[0021] Preferably, the process of constructing a data set for training a rare earth element concentration detection model includes:

[0022] S11: Control the ROV's robotic arm to advance in a zigzag pattern parallel to the seafloor topography, setting sampling points at equal intervals along the path to ensure uniform horizontal and vertical coverage of the entire target area. Furthermore, after each sampling step, the robotic arm is cleaned and reset before proceeding to the next sampling point.

[0023] S12, based on the sampling path, the rare earth enriched liquid sample of each sampling point is obtained by using the integrated deep-sea LIBS sampling device, and the spectral data of a total of N samples Sp1, Sp2, ..., Sp are obtained by performing laser excitation experiments on the underwater ROV. N ; At the same time, record the spatial information St of each sampling point i , including longitude lg i , latitude lt i , depth dp i , namely St i =[lg i ,lt i , dp i ], and i∈[1, N]; finally, we get the spatial information St1, St2, ..., St of N samples. N ;

[0024] S13, determine the type of rare earth element to be detected; then send the collected samples to the shipboard laboratory for concentration calibration experiment, and use inductively coupled plasma mass spectrometry to measure the rare earth element concentration values Rc1, Rc2, ..., Rc at N sampling points. N , which is used as the concentration label of the sample, where each rare earth element concentration value Rc i They all contain the concentration values of ten elements, and i∈[1,N];

[0025] S14, the acquired multi-point spectral data Sp1, Sp2, ..., Sp N and spatial position information St1, St2, ..., St N As input data, the rare earth element concentration values Rc1, Rc2, ..., Rc N As target data, the input data and target data are integrated as a complete rare earth detection data set.

[0026] Preferably, the specific process of S2 is as follows:

[0027] Graph node construction: Each sampling point is regarded as a node Node in the graph, and there are a total of N nodes. The LIBS spectral data corresponding to each node and the spatial position information are jointly used as the feature representation of the node; the feature vector of each node is represented as X i , and i ∈ [1, N];

[0028] Graph edge construction: Based on the spatial distance and spectral similarity between each pair of nodes, it is jointly determined whether to establish an edge connection; the three-dimensional spatial distance calculation between each pair of sampling points is represented as Dis ij , j ∈ [1, N] and j ≠ i;

[0029] Calculate the spectral similarity Cos between spectra ij , representing the spectral similarity between the i-th sampling point and the j-th sampling point. Cos ij ∈ [0, 1], and the larger Cos ij is, the higher the spectral similarity between the two sampling points. Conversely, the spectral similarity is lower; then, calculate the overall similarity Sim between the i-th sampling point and the j-th sampling point ij ;

[0030] Finally, select the K sampling points with the highest similarity to establish an edge Edge connection, and assign a feature vector containing spatial and spectral information to Edge:

[0031] E ij = [Dis ij , Sim ij

[0032] Among them, E ij is the Edge feature between node i and node j, and i, j ∈ [1, N];

[0033] Complete graph construction: Based on graph node construction and graph edge construction, the data containing spectral data Sp1, Sp2,..., Sp N and spatial position information St1, St2,..., St N is characterized as a complete graph representation G = (X, E), where X represents the set containing all N sampling point nodes, and each node is attached with a spectrum-position fusion feature vector X i ; E represents the set of all edges, and each edge is attached with a corresponding Edge feature vector E ij .

[0034] Preferably, the rare earth element concentration detection model based on the graph neural network GNN includes an input feature encoding layer, a spatial perception module, and a rare earth element prediction module;

[0035] The input feature encoding layer is used to take a graph representation G = (X, E) containing the node feature vector X i and the Edge feature vector E ij as input, and respectively perform feature encoding on X i and E ij through a multi-layer perceptron, and obtain the encoded node features and edge features

[0036] The spatial perception module sequentially inputs each node feature and edge feature output by the feature encoding layer into L spatial perception graph attention layers, and finally obtains the encoded features of each node where i ∈ [1, N], and uses them as the output features of the spatial perception module; the spatial perception module is composed of L spatial perception graph attention layers; in each layer, the attention weights are calculated by combining the similarity between node features and the spatial distance attenuation factor; through multi-layer spatial perception graph attention propagation, the model learns the context information of different neighborhood scales layer by layer;

[0037] The rare earth element prediction module, based on the obtained encoded features of each node performs feature compression on the node encoded features through a multi-layer perceptron, and then obtains the rare earth element concentration prediction results of each sampling point through a fully connected layer and a ReLU activation function

[0038] Preferably, the specific process of S4 is as follows:

[0039] Divide the overall area to be detected into R square grids of equal size, and divide each sampling point into different local area grids;

[0040] For each delimited area, summarize the rare earth element concentration prediction values of all sampling points within the area, and calculate the average concentration μ r and variance Δ r , μ r and Δ r respectively represent the rare earth average concentration and variance of the r-th area, and r ∈ [1, R];

[0041] Set multi-level discrimination rules and set threshold parameters, including the rare earth high enrichment mean parameter μ high , high enrichment variance parameter Δ high , and medium enrichment mean parameter μ mid ; use these threshold parameters to determine whether the area is a rare earth enrichment area and assign a rare earth enrichment level label.

[0042] In the second aspect of the present invention, a deep-sea LIBS rare earth element quantitative detection system integrating spatial information is provided, including a water surface part and an underwater part;

[0043] The underwater part includes a main cabin and an integrated deep-sea LIBS sampling device; inside the main cabin, there are a laser control module, a spectrometer, a power supply module, a control module, and an in-cabin status monitoring and feedback unit;

[0044] The LIBS optical probe is electrically connected to the spectrometer and includes a laser head and an auxiliary sampling device. The laser head is electrically connected to the laser control module; the integrated deep-sea LIBS sampling device is based on a regional sampling path and a point distribution strategy to perform multi-point deep-sea sediment sampling in the target area, obtain the sample to be measured, and through the auxiliary sampling device, perform excitation experiments through the laser head to obtain spectral characteristics;

[0045] The water surface part includes a control computer, which integrates a rare earth element concentration detection model based on the graph neural network GNN that has been constructed and trained with modular packaging, a spectral-position fusion graph structure data modeling module, and a regional aggregation algorithm module;

[0046] The spectral-position fusion graph structure data modeling module uses different sampling points as nodes Node in the graph structure, combines the spectral characteristics and spatial distances of the sampling points to construct the edge Edge structure of the graph, encodes the spectral characteristics and spatial position information of each sampling point as node features, and converts the sample data to be measured into graph structure data as the standard input data of the rare earth element concentration detection model;

[0047] The rare earth element concentration detection model extracts the spatial correlation characteristics of rare earth elements through the node feature propagation and aggregation mechanism of GNN and outputs the rare earth element concentration of each sampling point;

[0048] The regional aggregation algorithm module uses the regional aggregation algorithm based on the rare earth element concentration of each sampling point in the target area to calculate the regional-level rare earth enrichment trend and output the enrichment level.

[0049] Preferably, the specific structure of the integrated deep-sea LIBS sampling device is:

[0050] The device includes a sample cabin, a reaction chamber, an ROV manipulator, four groups of deep-water motors and their corresponding piston structures;

[0051] Among them, the fourth deep - water motor is connected to the sampling piston through a slider and is controlled by a robotic arm; the third deep - water motor is connected to the acid - liquid piston through a slider. The right chamber of the acid - liquid piston is connected to the reaction chamber through a pipeline, and an acid - adding port is provided in the connecting chamber; a stirring and drainage integrated pump assembly is arranged at the bottom of the reaction chamber; the reaction chamber is connected to the sample chamber through a pipeline and a stop valve; among them, the first deep - water motor is connected to the suction - filtration and decompression piston through a slider. The right chamber of the suction - filtration and decompression piston is connected to the sample chamber through a pipeline, and a pressure sensor is arranged on the sample chamber; among them, the second deep - water motor is connected to the seawater piston through a slider. The right chamber of the seawater piston is connected to the left chamber of the suction - filtration and decompression piston through a pipeline; a seawater inlet is arranged in the left chamber of the seawater piston.

[0052] Preferably, the rare - earth element concentration detection model based on the graph neural network GNN includes an input feature encoding layer, a spatial perception module, and a rare - earth element prediction module;

[0053] The input feature encoding layer is used to take a graph representation G=(X, E) containing node feature vectors X i and Edge feature vectors E ij as input, and respectively perform feature encoding on X i and E ij through a multi - layer perceptron, and obtain the encoded node features and edge features

[0054] The spatial perception module sequentially inputs each node feature and edge feature output by the feature encoding layer into L spatial perception graph attention layers, and finally obtains the encoded feature of each node where i ∈ [1, N], and uses it as the output feature of the spatial perception module; the spatial perception module is composed of L spatial perception graph attention layers; in each layer, the attention weight is calculated by combining the similarity between node features and the spatial distance attenuation factor; through multi - layer spatial perception graph attention propagation, the model gradually learns the context information of different neighborhood scales;

[0055] The rare - earth element prediction module, based on the obtained encoded feature of each node compresses the node encoded feature through a multi - layer perceptron, and then obtains the rare - earth element concentration prediction result of each sampling point through a fully - connected layer and a ReLU activation function

[0056] Compared with the prior art, the present invention has the following innovative points:

[0057] (1) Integrated Deep-sea Laser Induced Breakdown Spectroscopy (LIBS) Sampling Device: An integrated deep-sea LIBS device with functions of sediment sampling, acid dissolution in the reaction chamber, particle filtration, and electric field enrichment is proposed, which can achieve in-situ extraction and signal enhancement of rare earth elements;

[0058] (2) Spectral-Positional Fusion Graph Structure Data Modeling Method: A "spectral-positional fusion" graph modeling method for deep-sea rare earth element detection is constructed. The spectral features and spatial position information are jointly encoded as graph node features, and an edge structure considering spatial distance and spectral similarity is constructed to realize the expression of the distribution relationship of rare earth elements in complex geological environments;

[0059] (3) GNN-based Quantitative Detection Model and Region Aggregation Method for Rare Earth Elements: A graph neural network detection model combined with spectral-positional graph input is constructed. The context information between sampling points is extracted through a spatial-aware graph attention layer, and a region aggregation mechanism is introduced. Finally, accurate detection and identification of rare earth elements at the sampling point and regional levels are achieved.

[0060] The beneficial effects brought by the innovation points of the present invention include:

[0061] Realize in-situ quantitative detection in the deep-sea environment: An integrated deep-sea LIBS sampling and analysis device is adopted, which has high-pressure corrosion resistance and automatic sampling functions. Combined with the constructed graph neural network model, sample collection, signal collection, concentration analysis, and enrichment area identification can be completed on-site at the seabed, breaking through the limitations of traditional laboratory analysis processes and realizing an integrated sampling detection and identification process;

[0062] Enhance the enrichment area identification ability and regional adaptability: The model introduces the fusion expression of spatial proximity and spectroscopic features during modeling. It can not only accurately detect single-point samples but also intelligently identify the rare earth enrichment trend in a specific area, with excellent regional adaptability and identification ability;

[0063] Improve the efficiency and intelligent level of resource exploration: The present invention realizes high-degree integration and intelligent deployment at the hardware and algorithm levels, significantly improving the efficiency and automation level of deep-sea rare earth resource exploration, and providing technical support for subsequent resource evaluation and mining decision-making. Brief Description of the Drawings

[0064] Figure 1 It is the overall implementation logic flowchart of the detection method of the present invention.

[0065] Figure 2 It is the structure diagram of the integrated deep-sea LIBS sampling device of the present invention.

[0066] Figure 3 It is the structure diagram of the GNN-based rare earth element concentration detection model of the present invention.

[0067] Figure 4 This is the architecture diagram of the rare earth element quantitative detection system of the present invention.

[0068] Figure 5 This is the heat map of the average concentration detection of rare earth elements in the embodiment of the present invention.

[0069] Figure 6 This is the comparison result diagram of the detection accuracy of rare earth element concentrations at the sampling point level in the embodiment of the present invention.

[0070] Figure 7 This is the comparison result diagram of the detection accuracy of rare earth element concentrations at the regional level in the embodiment of the present invention. Specific implementation manners

[0071] The overall implementation logic of the present invention is as Figure 1 shown. The present invention proposes a method and system for quantitative detection of rare earth elements in deep sea LIBS integrating spatial information. The main process is as follows: First, design and construct an integrated deep sea LIBS sampling device, and realize in-situ sampling of deep sea sediments through a remotely operated vehicle (ROV), and integrate functions such as extraction, acid dissolution, filtration and enrichment, effectively improving the excitation and detection signal quality of rare earth elements; Subsequently, carry out multi-point sampling experiments according to a specific deep sea area sampling strategy to obtain a rare earth detection data set containing spectral characteristics, spatial position information and concentration labels; Then, taking the sampling points as the nodes of the graph structure, fuse the spectral data and spatial information to construct a spectrum-position fused graph structure data; After that, construct a quantitative detection model of rare earth elements based on the graph neural network (GNN), use the information propagation mechanism in the graph structure to capture the internal law of the spatial distribution of rare earth elements, and combine the regional aggregation method to realize the identification of rare earth enrichment areas; Finally, encapsulate and deploy the sampling device and the detection model to realize in-situ detection of rare earth elements in the deep sea environment and real-time identification of enrichment areas, providing intelligent and regional technical support for deep sea resource exploration.

[0072] The following further illustrates the invention with specific embodiments.

[0073] The implementation process of the method for quantitative detection of rare earth elements in deep sea LIBS integrating spatial information in this embodiment is as follows:

[0074] S1. Design an integrated deep sea laser-induced breakdown spectroscopy (LIBS) sampling device; realize in-situ collection of deep sea sediment samples through the operation of a remotely operated vehicle (ROV), wherein the sampling device has functions of sediment extraction, acid dissolution in the reaction chamber, particle filtration and electric field enrichment for extracting rare earth element samples to be measured;

[0075] S2. Design a sampling strategy for the deep - sea area and construct a dataset; Based on the S1 integrated deep - sea LIBS sampling device, design the regional sampling path and point distribution strategy, obtain multi - point rare - earth element samples to be measured covering the target area, conduct excitation experiments through the LIBS optical probe to obtain spectral data, and based on the sampling path and point distribution strategy, obtain and record the corresponding spatial position information (including longitude, latitude, and depth), and obtain the rare - earth element concentration labels through laboratory calibration, thereby constructing a rare - earth detection dataset containing spectral characteristics, spatial information, and concentration labels;

[0076] S3. Spectral - position fusion graph - structure data modeling; Use different sampling points as nodes in the graph structure, combine the spectral characteristics and spatial distances of the sampling points to construct the edge structure of the graph, and encode the spectral characteristics and spatial position information of each sampling point as node features, thereby converting the dataset in S2 into graph - structure data;

[0077] S4. Construct a rare - earth element concentration detection model based on the graph neural network (GNN); Use the graph - structure data described in S3, extract the spatial correlation features of rare - earth elements through the node feature propagation and aggregation mechanism of GNN, realize the quantitative detection of the rare - earth element concentration at each sampling point, and at the same time combine the regional aggregation method to realize the identification and detection of the regional - level rare - earth enrichment trend;

[0078] S5. System deployment and application; Integrate, package, and deploy the rare - earth element concentration detection model to realize in - situ quantitative detection of rare - earth elements and identification of enrichment areas in the deep - sea environment, sample in real - time and output the rare - earth concentration prediction results at the sampling - point level and regional level.

[0079] I. Specific implementation of the integrated deep - sea LIBS sampling device

[0080] The components in deep - sea sediments are complex and the environment is extremely harsh. To exclude the influence of other components in the sediments on the rare - earth element LIBS signal and enhance the excitation and collection efficiency of the rare - earth element LIBS signal, the present invention designs an integrated deep - sea LIBS sampling device, which can filter out other particulate matters in the sediments and enrich rare - earth elements through electric - field phase transformation, and can finally realize the collection of rare - earth elements in the sediments and the enhancement of LIBS signals. The sampling device is as Figure 2 shown:

[0081] The device includes a sample chamber, a reaction chamber, an ROV manipulator, four groups of deep - water motors and their corresponding piston structures;

[0082] Among them, the fourth deep - water motor is connected to the sampling piston through a slider and is controlled by a robotic arm; the third deep - water motor is connected to the acid piston through a slider. The right chamber of the acid piston is connected to the reaction chamber through a pipeline, and an acid - adding port is arranged in the connecting chamber; a stirring and drainage integrated pump assembly is arranged at the bottom of the reaction chamber; the reaction chamber is connected to the sample chamber through a pipeline and a stop valve; among them, the first deep - water motor is connected to the suction filtration and decompression piston through a slider. The right chamber of the suction filtration and decompression piston is connected to the sample chamber through a pipeline, and a pressure sensor is arranged on the sample chamber; among them, the second deep - water motor is connected to the seawater piston through a slider. The right chamber of the seawater piston is connected to the left chamber of the suction filtration and decompression piston through a pipeline; a seawater inlet is arranged in the left chamber of the seawater piston.

[0083] Sampling process: The purpose of this process is to complete the extraction, acid dissolution, stirring and concentration pretreatment of deep - sea sediments, and finally obtain rare - earth element samples for LIBS detection. The specific steps are as follows:

[0084] (1) Sediment sampling: Through the operation of the ROV robotic arm, the sampling piston driven by the fourth deep - water motor is inserted into the deep - sea sediments in the predetermined area. The motor drives the piston to draw backward to complete the collection of sediment samples; then the piston is inserted into the reaction chamber and driven in the reverse direction to push the sediment into the reaction chamber.

[0085] (2) Automatic sealing of the reaction chamber and acid injection: After pulling out the piston, the spring cover on the reaction chamber automatically closes to ensure the airtightness of the chamber environment; the third deep - water motor drives the acid piston to inject a preset volume of quantitative acid into the reaction chamber for sample pretreatment.

[0086] (3) Sample dissolution and mixing: Start the stirring and drainage integrated pump to drive the impeller in the chamber to rotate at high speed, so that the acid and the sediment are fully mixed and reacted, and the rare - earth elements in the sediment are fully dissolved into the acid to achieve component release.

[0087] (4) Sample transfer: After the mixing reaction is completed, the first deep - water motor drives the piston to extract the treated mixed sample in the reaction chamber into the special sample chamber for LIBS detection.

[0088] (5) Pressure regulation of the sample chamber: After closing the electric stop valve of the sample chamber, the deep - water motor continues to extract. Due to the system sealing, the pressure in the chamber will gradually decrease. The pressure sensor monitors the pressure of the sample chamber. When the pressure of the sample chamber drops to the set value (set to 11 MPa in the present invention), the sample preparation is completed.

[0089] (6) Obtaining the sample to be measured: At this time, the sample is already in the target pressure and sealed state. As the sample S to be measured for LIBS detection, it can be directly used for the laser excitation experiment.

[0090] Device Cleaning and Circulation Reset Process: The purpose of this process is to complete the cleaning and reset of the sample chamber and the pipelines in the system, prepare for the next sampling cycle, and thus ensure the long-term operation stability of the system. It specifically includes the following steps:

[0091] (1) Preparation for Seawater Flushing after Sampling: After the detection is completed, the second deep-sea motor drives the seawater piston to pump a preset volume of clean seawater; then the stop valve is opened, and the piston is further pushed to cross the flushing interface position;

[0092] (2) Flushing of the Sample Chamber: The first deep-sea motor continues to drive the seawater piston to push out all the seawater in the piston, completing the flushing of the sample chamber to prevent residual contamination from affecting the next detection;

[0093] (3) Cleaning of the Reaction Chamber: Synchronously start the reverse function of the stirring and drainage integrated pump to discharge the residual liquid and samples in the reaction chamber out of the chamber, completing the cleaning of the chamber;

[0094] (4) Device Reset: All motors return to their positions and standby, and the device is ready for the next sampling task.

[0095] Therefore, the sampling process using the integrated deep-sea LIBS sampling device can obtain the sample S to be detected, and then prepare for the next sampling process through the cleaning reset and preparation process. By using the integrated deep-sea LIBS sampling device, first execute the sampling process to effectively obtain the rare earth element sample to be detected; subsequently, execute the device cleaning and circulation reset process to thoroughly flush and reset the sample chamber and the pipeline system, and prepare for the next sampling process.

[0096] II. Design of Sampling Strategy for Deep-Sea Areas and Construction of Rare Earth Detection Dataset

[0097] First, based on the integrated deep-sea LIBS sampling device, formulate a regional sampling strategy suitable for the deep-sea environment; combine the sampling process and the device cleaning and circulation reset process to carry out multi-point deep-sea sediment sampling experiments, so as to obtain multi-point LIBS spectral data and corresponding spatial position information (including longitude, latitude, and depth) of the target area; then obtain the rare earth element concentration labels through laboratory calibration to construct a rare earth detection dataset. This process includes:

[0098] S2-1, Control the ROV manipulator to move forward along a Z-shaped route parallel to the seabed topography, and set sampling points at equal intervals on the path to ensure uniform coverage of the entire target area horizontally and vertically; in addition, perform the cleaning and reset operation after each sampling, and then go to the next sampling point;

[0099] S2-2. Based on the ROV sampling path of S2-1, use the LIBS sampling device to execute the sampling process, the device cleaning and cycle reset process, respectively obtain rare earth enriched liquid samples at each sampling point, and obtain spectral data Sp1, Sp2,..., Sp of a total of N samples through laser excitation experiments performed underwater by the ROV. N Meanwhile, record the spatial information St of each sampling point. i including longitude lg i latitude lt i depth dp i That is, St i = [lg i , lt i , dp i , and i ∈ [1, N]; finally, obtain the spatial information St1, St2,..., St of N samples. N ;

[0100] S2-3. Determine that the rare earth categories to be detected include ten types: lanthanum (La), cerium (Ce), praseodymium (Pr), neodymium (Nd), samarium (Sm), europium (Eu), gadolinium (Gd), terbium (Tb), dysprosium (Dy), holmium (Ho); then, synchronously send the collected samples to the on-board laboratory for concentration calibration experiments, and use inductively coupled plasma mass spectrometry (ICP-MS) to measure the rare earth element concentration values Rc1, Rc2,..., Rc of N sampling points. N Take them as the concentration labels of the samples, where each rare earth element concentration value Rc i contains the concentration values of ten elements, and i ∈ [1, N];

[0101] S2-4. Construct a data set, use the multi-point spectral data Sp1, Sp2,..., Sp N and spatial position information St1, St2,..., St N obtained in the above steps as input data, and the rare earth element concentration values Rc1, Rc2,..., Rc N as target data, and integrate the input data and target data as a complete rare earth detection data set;

[0102] S2-5. Repeat steps S2-1 to S2-5, so as to obtain a total of M groups of rare earth detection data sets by sampling in different regions.

[0103] III. Graph structure data modeling process for spectrum-position fusion

[0104] Based on the constructed rare earth detection dataset, in order to capture the spatial distribution characteristics of rare earth elements and mine the local correlation between spectral data, the present invention conducts graph structure data modeling of spectrum-position fusion; specifically, each sampling point is regarded as a node in the graph structure, and the edge connection relationship is constructed based on the spatial distance and spectral feature similarity between nodes, and multi-dimensional feature representations are assigned to the nodes and edges respectively to achieve graph structure data modeling of multi-source information fusion; its specific process includes:

[0105] Graph node construction: Each sampling point is regarded as a node in the graph and there are a total of N nodes, and the LIBS spectral data corresponding to each node and the spatial position information are jointly used as the feature representation of the node; each node feature vector can be expressed as:

[0106] X i =[Sp i ,St i =[Sp i ,lg i ,lt i ,dp i

[0107] Among them, X i represents the node feature vector of the i-th sampling point, Sp i ,St i are the spectral data and spatial position information of the i-th sampling point respectively, lg i ,lt i ,dp i are the longitude, latitude and depth positions of the i-th sampling point respectively, and i ∈ [1, N];

[0108] Graph edge construction: Based on the spatial distance and spectral similarity between each pair of nodes, it is jointly determined whether to establish an edge connection; the calculation of the three-dimensional spatial distance between each pair of sampling points is expressed as:

[0109]

[0110] Among them, Dis ij represents the spatial distance between the i-th sampling point and the k-th sampling point, lg j ,lt j ,dp j are the longitude, latitude and depth positions of the j-th sampling point respectively, k ∈ [1, N] and j ≠ i;

[0111] Calculate the spectral similarity between spectra:

[0112]

[0113] Among them, Cos ij ​represents the spectral similarity between the \(i\)-th sampling point and the \(k\)-th sampling point, Cos ij ∈[0, 1] and Cos ij The larger the value, the higher the spectral similarity between the two sampling points. Conversely, the lower the spectral similarity; Sp j represents the spectral data of the \(j\)-th sampling point, where \(j\in[1, N]\) and \(j\neq i\); ||*|| is the modulus operation;

[0114] After that, calculate the overall similarity Sim between the \(i\)-th sampling point and the \(k\)-th sampling point ij :

[0115]

[0116] where \(\alpha_1\) and \(\alpha_1\) are weight coefficients; for the \(i\)-th sampling point, calculate the overall similarity between this sampling point and other sampling points, Sim i1 , Sim i2 ,..., Sim iN , where Sim i1 and Sim iN represent the overall similarities between the \(i\)-th sampling point and the 1st and \(N\)-th sampling points respectively;

[0117] Finally, select the \(K\) sampling points with the highest similarity to establish an edge (Edge) connection, and assign a feature vector containing spatial and spectral information to Edge:

[0118] E ij = [Dis ij , Sim ij

[0119] where \(E\) ij is the Edge feature between node \(i\) and node \(j\), and \(i, k\in[1, N]\).

[0120] Complete graph construction: Based on graph node construction and graph edge construction, represent the spectral data Sp1, Sp2,..., Sp N and spatial position information St1, St2,..., St N in the dataset as a complete graph representation \(G=(X, E)\), where \(X\) represents the set containing all \(N\) sampling point nodes, and each node is attached with a spectrum-position fusion feature vector \(X\) i ; \(E\) represents the set of all edges, and each edge is attached with a corresponding Edge feature vector \(E\) ij ; Finally, the graph representation \(G=(X, E)\) is used as the input of the rare earth element quantitative detection model based on graph neural network and is used to predict the rare earth element concentration values Rc1, Rc2,..., Rc N . ​

[0121] IV. Rare Earth Element Concentration Detection Model Based on GNN

[0122] In order to fully explore the correlation of rare earth element concentrations in spatial distribution and improve the quantitative recognition ability of LIBS spectral data, the present invention constructs a rare earth element concentration detection model based on GNN; this model takes the constructed spectral-position fusion graph structure data as input, uses the feature propagation and aggregation mechanism of the graph neural network to learn the correlation relationship between nodes in the global learning graph structure from the global, and realizes the accurate prediction of the rare earth element concentration at each sampling point; at the same time, through the regional-level aggregation algorithm, the recognition and detection of the rare earth enrichment trend within the region are realized; the rare earth element concentration detection model based on GNN is as Figure 3 shown, and specifically includes the following steps:

[0123] Input Feature Encoding Layer: Taking the graph representation G=(X, E) containing the node feature vector X i and the Edge feature vector E ij as input, respectively perform feature encoding on X i and E ij through a multi-layer perceptron, and obtain the encoded node feature and edge feature

[0124] Spatial Perception Module: This module performs spatial correlation modeling and feature interaction on the encoded node feature and edge feature to achieve spatial structure constraint modeling; this module consists of L spatial perception graph attention layers; in each layer, calculate the attention weight by combining the similarity between node features and the spatial distance attenuation factor, ensure that adjacent nodes with high similarity obtain higher weights, and suppress the interference of distant noise nodes on the central node representation, thereby improving the accuracy of local spatial modeling; finally, through multi-layer spatial perception graph attention propagation, the model can gradually learn the context information of different neighborhood scales and realize the in-depth exploration of the distribution law of rare earth elements; the calculation of the spatial perception graph attention layer is specifically expressed as:

[0125]

[0126] where and respectively represent the output and input node features of the i-th node in the l-th layer of the spatial perception graph attention layer, and l∈[1, L]; N(i) represents the set of neighbor nodes of the i-th node, ReL∪(*) is the ReLU activation function, ω ij represents the spatial attention weight between the i-th node and the j-th node; is the basic attention weight between the i-th node and the j-th node, βij is the corresponding spatial attenuation factor, and β ij is calculated as:

[0127]

[0128] where, and respectively represent the feature representations of node i and node j at the l-th layer. When l = 0, it is the output feature of the feature encoding layer and Dis ij is the spatial distance between the i-th sampling point and the j-th sampling point, and σ is the spatial attenuation factor parameter;

[0129] Therefore, each node feature and edge feature output by the feature encoding layer are sequentially input into L spatial-aware graph attention layers, and finally the encoded features of each node are obtained where i ∈ [1, N], and it is used as the output feature of the spatial-aware module.

[0130] Rare earth element prediction module: Based on the obtained encoded features of each node The encoded features of the nodes are compressed through a multi-layer perceptron, and then the rare earth element concentration prediction results of each sampling point are obtained through a fully connected layer and a ReLU activation function

[0131] Regional-level aggregation algorithm: Based on the rare earth element concentration prediction results of each sampling point To realize the recognition of the regional-level rare earth element distribution trend and accurately judge and divide the regional enrichment degree, the present invention designs a regional-level aggregation algorithm by combining spatial information, which specifically includes the following steps:

[0132] (1) Divide the overall area to be detected into R equal-sized square grids, and divide each sampling point into different local area grids;

[0133] (2) For each delimited area, sum up the rare earth element concentration prediction values of all sampling points in the area, and calculate the average concentration μ r and variance Δ r , μ r and Δ r respectively represent the rare earth average concentration and variance of the r-th area, and r ∈ [1, R];

[0134] (3) Set multi-level discrimination rules and threshold parameters, including the rare earth high enrichment mean parameter μ high , high enrichment variance parameter Δ high , and medium enrichment mean parameter μ mid; These threshold parameters are used to determine whether the area is a rare earth enrichment area and assign rare earth enrichment level labels, as shown in Table 1:

[0135] Table 1 Multi-level discrimination rules for rare earth enrichment areas

[0136] Rare earth enrichment level Discrimination condition High enrichment area <![CDATA[μ r > μ high and Δ r < Δ high > Medium enrichment area <![CDATA[μ mid <m r <m high or r >m high ,D r >]]> Low enrichment area <![CDATA[μ r <μ mid >

[0137] V. Quantitative detection system based on rare earth element concentration detection model

[0138] Training of the rare earth element concentration detection model based on GNN: Based on the constructed rare earth detection data set and through the spectral-position fusion graph structure data modeling method, a standard model training data set is obtained, and the model is trained. During the training process, the mean square error (MSE) is used to calculate the loss function of element concentration prediction; Adam optimization algorithm is used for parameter update during model training, and finally the trained rare earth element concentration detection model is obtained.

[0139] The detection system is as Figure 4 shown, including the water surface part and the underwater part;

[0140] Among them, the underwater part includes the main cabin and the integrated deep-sea LIBS sampling device; the main cabin contains a laser control module, a spectrometer, a power supply module, a control module, and an in-cabin status monitoring and feedback unit;

[0141] The LIBS optical probe is electrically connected to the spectrometer and includes a laser head and an auxiliary sampling device. The laser head is electrically connected to the laser control module; the integrated deep-sea LIBS sampling device is based on the regional sampling path and point distribution strategy, conducts multi-point deep-sea sediment sampling in the target area to obtain samples to be measured, and through the auxiliary sampling device, conducts excitation experiments through the laser head to obtain spectral characteristics;

[0142] Among them, the water surface part includes a control computer, which integrates a modularly packaged rare earth element concentration detection model based on the graph neural network GNN, a spectral-position fusion graph structure data modeling module, and a regional aggregation algorithm module that have been constructed and trained;

[0143] The spectral-position fusion graph structure data modeling module uses different sampling points as nodes Node in the graph structure, combines the spectral characteristics and spatial distances of the sampling points to construct the edge Edge structure of the graph, encodes the spectral characteristics and spatial position information of each sampling point as node features, and converts the sample data to be measured into graph structure data as the standard input data of the rare earth element concentration detection model;

[0144] The rare earth element concentration detection model extracts the spatial correlation characteristics of rare earth elements through the node feature propagation and aggregation mechanism of GNN and outputs the rare earth element concentration of each sampling point;

[0145] The regional aggregation algorithm module uses the regional aggregation algorithm based on the rare earth element concentrations of each sampling point in the target area to calculate the regional-level rare earth enrichment trend and output the enrichment level.

[0146] Based on the obtained rare earth element concentration distribution and enrichment level results, the staff combines geological background information and actual exploration requirements to optimize the target area and deploy further operation decisions.

[0147] VI. Explanation of Experimental Results

[0148] To verify the effectiveness of the proposed deep-sea LIBS rare earth element quantitative detection method in rare earth element concentration prediction, the present invention conducted experimental tests and obtained the average concentration distributions of ten rare earth elements (La, Ce, Pr, Nd, Sm, Eu, Gd, Tb, Dy, Ho) in different regions, as Figure 5 shown; the results of the heat map analysis indicate that the method of the present invention can effectively achieve regional-level detection of the spatial distribution of rare earth elements, further verifying the feasibility of the method in deep-sea rare earth element quantitative analysis.

[0149] In addition, the present invention conducted a comparative analysis with three existing mainstream algorithm models for rare earth element concentration prediction, including support vector machine (SVM), random forest (RF), and gradient boosting regression tree (GBRT). Moreover, taking the mean absolute percentage error (MAPE) as the evaluation index, the rare earth element concentration prediction accuracies of the models at the sampling point level and the regional level were compared; specifically, multiple sampling points were selected in the area to be detected, and the predicted values of the rare earth element concentrations of each sampling point were obtained, and the average of all sampling points was taken and compared with the true element concentration to calculate the MAPE to obtain the final detection accuracy; similarly, for the evaluation of the rare earth element concentration prediction accuracy at the regional level, the entire area was first divided into grids, and the average of the rare earth detection concentrations of the sampling points within the area was taken, and the MAPE was calculated to obtain the rare earth concentration detection accuracy within the area; the experimental results are as Figure 6 and Figure 7 shown;

[0150] Because the method of the present invention uses an integrated deep-sea LIBS sampling device for rare earth sampling, it can effectively alleviate the interference caused by spectral overlap and complex soil composition, thereby improving the discrimination and quantitative ability of rare earth element identification. In addition, this method introduces a spatial attenuation attention factor based on GNN, which can fully utilize the spatial correlation of the distribution of rare earth elements in the geological background. By jointly modeling the spectral characteristics of adjacent sampling points with the spatial information of the target point, the model's ability to distinguish subtle differences is ultimately improved. Therefore, compared with existing methods (support vector machine (SVM), random forest (RF), and gradient boosted regression tree (GBRT), the method of the present invention shows higher prediction accuracy in the detection of ten rare earth elements, reflecting stronger feature extraction capabilities and refined modeling level.

[0151] Furthermore, the proposed method enhances the model's robustness to local anomalies by introducing a graph modeling approach based on spatial information, thereby avoiding the impact of single-point data fluctuations on the overall prediction results. Furthermore, the GNN's propagation mechanism on graph data can deeply capture the potential connections between different sampling points, thereby enhancing the model's generalization ability across different geological regions and significantly reducing performance fluctuations in different element detection applications. Experimental results further validate the stability of the proposed method in detecting the concentrations of ten rare earth elements, thereby ensuring reliable prediction performance with high accuracy and low fluctuations even under complex geological conditions.

[0152] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

[0153] Although the above describes the specific implementation methods of the present invention, it does not limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for quantitatively detecting rare earth elements in deep sea LIBS by integrating spatial information, characterized in that, It includes the following processes: S1. Based on the integrated deep-sea LIBS sampling device and the regional sampling path and point distribution strategy, multi-point deep-sea sediment sampling is carried out in the target area to obtain the rare earth element samples to be measured in the target area, and the corresponding spatial position information is recorded. Then, the spectral characteristics of each sample to be measured are detected by laser excitation; S2. Using different sampling points as nodes Node in the graph structure, combining the spectral characteristics and spatial distances of the sampling points to construct the edge Edge structure of the graph, and encoding the spectral characteristics and spatial position information of each sampling point as node features, converting the sample data to be measured into graph structure data, which is used as the standard input data for the rare earth element concentration detection model; S3. Input the graph structure data in S2 into the constructed and trained rare earth element concentration detection model based on the graph neural network GNN. The model extracts the spatial correlation features of rare earth elements through the node feature propagation and aggregation mechanism of GNN, and outputs the rare earth element concentration of each sampling point; S4. Based on the rare earth element concentration of each sampling point in the target area, use the regional aggregation algorithm to calculate the regional-level rare earth enrichment trend and output the enrichment level.

2. The quantitative detection method for rare earth elements in deep - sea LIBS integrating spatial information according to claim 1, wherein, The specific structure and sampling process of the integrated deep-sea LIBS sampling device are as follows: The device includes a sample chamber, a reaction chamber, an ROV manipulator, four groups of deep-water motors and their corresponding piston structures; Among them, the fourth deep-water motor is connected to the sampling piston through a slider and is controlled by the manipulator; the third deep-water motor is connected to the acid piston through a slider. The right chamber of the acid piston is connected to the reaction chamber through a pipeline, and an acid addition port is arranged in the connecting chamber; a stirring and drainage integrated pump assembly is arranged at the bottom of the reaction chamber; the reaction chamber is connected to the sample chamber through a pipeline and a stop valve; among them, the first deep-water motor is connected to the suction filtration and pressure reduction piston through a slider, the right chamber of the suction filtration and pressure reduction piston is connected to the sample chamber through a pipeline, and a pressure sensor is arranged on the sample chamber; the second deep-water motor is connected to the seawater piston through a slider, and the right chamber of the seawater piston is connected to the left chamber of the suction filtration and pressure reduction piston; a seawater inlet is arranged in the left chamber of the seawater piston; Through the operation of the ROV manipulator, the sampling piston driven by the fourth deep-water motor is inserted into the deep-sea sediment in the predetermined area, and the motor drives the piston to draw backward to complete the collection of the sediment sample; then the piston is inserted into the reaction chamber and driven in the reverse direction to push the sediment into the reaction chamber; The third deep-water motor drives the acid piston to inject a preset volume of quantitative acid into the reaction chamber for sample pretreatment; Start the stirring and drainage integrated pump to drive the impeller in the chamber to rotate, so that the acid and the sediment are fully mixed and reacted, and the rare earth elements in the sediment are fully dissolved in the acid to achieve component release; After the mixing reaction is completed, the first deep-water motor drives the piston to extract the processed mixed sample in the reaction chamber into the special sample chamber for LIBS detection; After closing the electric stop valve of the sample chamber, the first deep-water motor continues to extract. Due to the system seal, the pressure in the chamber will gradually decrease. The pressure sensor monitors the pressure of the sample chamber. When the pressure of the sample chamber drops to the set value, the sample is in the target pressure and sealed state, serving as the sample to be measured for LIBS detection and can be directly used for laser excitation experiments.

3. The quantitative detection method for rare earth elements in deep - sea LIBS integrating spatial information according to claim 1, characterized in that, The construction process of the dataset for training the rare earth element concentration detection model includes: S11. Control the ROV manipulator to move forward along a zigzag route parallel to the seabed topography, and set sampling points at equal intervals on the path to ensure uniform coverage of the entire target area both horizontally and vertically. In addition, after each sampling, perform a cleaning and reset operation and then move to the next sampling point; S12. Based on the sampling path, use the integrated deep-sea LIBS sampling device to obtain rare-earth enriched liquid samples at each sampling point respectively, and obtain the spectral data Sp1, Sp2,..., Sp of a total of N samples through laser excitation experiments carried out by an underwater ROV N ; Meanwhile, record the spatial information St i of each sampling point, including longitude lg i , latitude lt i , and depth dp i , that is, St i = [lg i , lt i , dp i , and i ∈ [1, N]; Finally, obtain the spatial information St1, St2,..., St of N samples N ; S13. Determine the rare earth category to be detected; then synchronously send the collected samples to the on-board laboratory for concentration calibration experiments, and use inductively coupled plasma mass spectrometry to measure the rare earth element concentration values Rc1, Rc2,..., RCN at N sampling points, and use them as the concentration labels of the samples, where each rare earth element concentration value Rc N contains the concentration values of ten elements, and i ∈ [1, N]; i ​ S14, take the obtained multi-point spectral data Sp1, Sp2,..., Sp N and spatial position information St1, St2,..., St N as input data, and the rare earth element concentration values Rc1, Rc2,..., Rc N as target data, and integrate the input data and the target data as a complete rare earth detection data set.

4. The quantitative detection method for rare earth elements by deep-sea LIBS integrating spatial information according to claim 1, wherein: The specific process of S2 is as follows: Graph Node Construction: Each sampling point is regarded as a node Node in the graph, and there are a total of N nodes. The LIBS spectral data corresponding to each node and the spatial position information are jointly used as the feature representation of the node; the feature vector of each node is represented as X i , and i ∈ [1, N]; Graph edge construction: Based on the spatial distance and spectral similarity between each pair of nodes, jointly determine whether to establish an edge connection; the calculation of the three-dimensional spatial distance between each pair of sampling points is expressed as Disi j , j ∈ [1, N] and j ≠ i; Calculate the spectral similarity Cos between spectra ij , representing the spectral similarity between the i-th sampling point and the j-th sampling point, Cos ij ∈ [0, 1] and the larger Cos ij is, the higher the spectral similarity between the two sampling points. Conversely, the lower the spectral similarity; then, calculate the overall similarity Sim ij ; Finally, select the K sampling points with the highest similarity to establish an Edge connection, and assign a feature vector containing spatial and spectral information to the Edge: E ij = [Dis ij , Sim ij ​ Among them, E ij is the Edge feature between node i and node j, where i, j ∈ [1, N]; Complete graph construction: Based on graph node construction and graph edge construction, it represents the data including spectral data Sp1, Sp2, ..., Sp N and spatial position information St1, St2, ..., St N as a complete graph representation G = (X, E), where X represents the set containing all N sampling point nodes, and each node is attached with a spectrum-position fusion feature vector X i ; E represents the set of all edges, and each edge is attached with a corresponding Edge feature vector E ij .

5. A deep-sea LIBS rare earth element quantitative detection method integrating spatial information according to claim 1, characterized in that: The rare earth element concentration detection model based on the graph neural network GNN includes an input feature encoding layer, a spatial perception module, and a rare earth element prediction module; The input feature encoding layer is used to take a graph representation G=(X, E) containing node feature vector X i and Edge feature vector E ij as input, and respectively perform feature encoding on X i and E ij through a multi-layer perceptron, and obtain the encoded node features and edge features The spatial perception module sequentially inputs each node feature and edge feature output by the feature encoding layer into L spatial perception graph attention layers, and finally obtains the encoded features of each node. where i ∈ [1, N], and it is used as the output feature of the spatial perception module; the spatial perception module is composed of L spatial perception graph attention layers; in each layer, the attention weights are calculated by combining the similarity between node features and the spatial distance decay factor; through multi-layer spatial perception graph attention propagation, the model gradually learns the context information of different neighborhood scales. The rare earth element prediction module is based on the obtained encoded features of each node Feature compression is performed on the encoded features of the nodes through a multi-layer perceptron, and then the prediction results of the rare earth element concentrations at each sampling point are obtained through a fully connected layer and a ReLU activation function 6. The quantitative detection method for rare earth elements in deep-sea LIBS integrating spatial information according to claim 1, wherein: The specific process of S4 is as follows: Divide the overall area to be detected into R square grids of equal size, and divide each sampling point into different local area grids; For each delineated area, summarize the predicted values of the rare earth element concentrations at all sampling points within the area, and calculate the average concentration μ within the area r and the variance Δ r , μ r and Δ r represent the average rare earth concentration and variance of the r-th area respectively, and r ∈ [1, R]; Set multi-level discrimination rules and set threshold parameters, including the rare earth high enrichment mean parameter μ high , the high enrichment variance parameter Δ high , and the medium enrichment mean parameter μ mid ; Use these threshold parameters to determine whether the area is a rare earth enrichment area and assign rare earth enrichment level labels.

7. A deep-sea LIBS rare earth element quantitative detection system integrating spatial information, characterized in that: It includes a water surface part and an underwater part; Among them, the underwater part includes a main cabin and an integrated deep-sea LIBS sampling device; the main cabin contains a laser control module, a spectrometer, a power supply module, a control module, and an in-cabin status monitoring and feedback unit; The LIBS optical probe is electrically connected to the spectrometer and includes a laser head and an auxiliary sampling device. The laser head is electrically connected to the laser control module; the integrated deep-sea LIBS sampling device performs multi-point deep-sea sediment sampling in the target area based on the regional sampling path and point distribution strategy to obtain samples to be detected, and through the auxiliary sampling device, excitation experiments are carried out through the laser head to obtain spectral characteristics; Among them, the water surface part includes a control computer, which integrates a rare earth element concentration detection model based on the graph neural network GNN, a spectrum-position fusion graph structure data modeling module, and a regional aggregation algorithm module that are constructed and trained in a modular package; The spectrum-position fusion graph structure data modeling module uses different sampling points as nodes Node in the graph structure, combines the spectral characteristics and spatial distances of the sampling points to construct the Edge structure of the graph, encodes the spectral characteristics and spatial position information of each sampling point as node features, and converts the sample data to be detected into graph structure data as the standard input data of the rare earth element concentration detection model; The rare earth element concentration detection model extracts the spatial correlation characteristics of rare earth elements through the node feature propagation and aggregation mechanism of GNN and outputs the rare earth element concentration of each sampling point; The regional aggregation algorithm module calculates the regional-level rare earth enrichment trend based on the rare earth element concentration of each sampling point in the target area using the regional aggregation algorithm and outputs the enrichment level.

8. The deep-sea LIBS rare earth element quantitative detection system integrating spatial information according to claim 7, wherein: The specific structure of the integrated deep-sea LIBS sampling device is: The device includes a sample cabin, a reaction chamber, an ROV manipulator, four groups of deep-water motors and their corresponding piston structures; Among them, the fourth deep-water motor is connected to the sampling piston through a slider and is controlled by a robotic arm; the third deep-water motor is connected to the acid piston through a slider. The right chamber of the acid piston is connected to the reaction chamber through a pipeline, and an acid addition port is provided in the connecting chamber; a stirring and drainage integrated pump assembly is provided at the bottom of the reaction chamber; the reaction chamber is connected to the sample chamber through a pipeline and a stop valve; among them, the first deep-water motor is connected to the suction filtration and decompression piston through a slider, the right chamber of the suction filtration and decompression piston is connected to the sample chamber through a pipeline, and a pressure sensor is provided on the sample chamber; among them, the second deep-water motor is connected to the seawater piston through a slider, and the right chamber of the seawater piston is connected to the left chamber of the suction filtration and decompression piston through a pipeline; a seawater inlet is provided in the left chamber of the seawater piston.

9. The deep-sea LIBS rare-earth element quantitative detection system integrating spatial information according to claim 7, characterized in that: The rare earth element concentration detection model based on the graph neural network GNN includes an input feature encoding layer, a spatial perception module, and a rare earth element prediction module; The input feature encoding layer is used to take a graph representation G = (X, E) that includes node feature vector X i and Edge feature vector E ij as input, and respectively perform feature encoding on X i and E ij through a multi-layer perceptron, and obtain the encoded node features and edge features The spatial perception module sequentially inputs each node feature and edge feature output by the feature encoding layer into L spatial perception graph attention layers, and finally obtains the encoded features of each node. Where i ∈ [1, N], and it is used as the output feature of the spatial perception module; the spatial perception module consists of L spatial perception graph attention layers; in each layer, the attention weight is calculated by combining the similarity between node features and the spatial distance attenuation factor; through multi-layer spatial perception graph attention propagation, the model gradually learns the context information of different neighborhood scales. The rare earth element prediction module is based on the obtained encoding features of each node The encoding features of the nodes are compressed by a multi-layer perceptron, and then the prediction results of the rare earth element concentrations at each sampling point are obtained through a fully connected layer and a ReLU activation function

Images