Method and system for discriminating tectonic environment based on apatite geochemical characteristics

Abstract

The present application relates to the technical field of geochemistry and geological data analysis, and discloses a method and system for judging tectonic environment based on apatite geochemical characteristics, which comprises the following steps: obtaining a target sample containing a reference group and an evolution reserved group of particles, and saving the coordinates and parameters of the two in a database; performing in-situ micro-area analysis on the reference group to obtain geochemical composition data; preprocessing the composition data to construct a static feature vector; inputting the static feature vector into a classification model containing a geological origin penalty matrix in a loss function, and outputting an original posterior probability; calculating the difference between the probabilities of the maximum two categories, and when the difference is less than a confidence threshold, performing a line scanning test on the evolution reserved group of particles according to the coordinates, extracting spatial composition evolution gradient features, and outputting a final judgment result in combination with the feature vector. The present application triggers a line scanning strategy through a confidence difference to avoid comprehensive scanning tests, and takes into account the analysis efficiency and accuracy; and the use of a geological rule constraint algorithm model reduces the misjudgment rate.

Description

Technical Field

[0001] This invention relates to the field of geochemistry and geological data analysis technology, specifically to a method and system for identifying tectonic environments based on the geochemical characteristics of apatite. Background Technology

[0002] Apatite, a widely found accessory mineral in various igneous rocks, can record the characteristics of magma source regions and crystallization evolution processes through its trace element composition. With the development of in-situ micro-area analysis technology, the use of classification algorithms to process apatite geochemical data and determine the rock genesis and its tectonic environment has become an important means of geological exploration and geochemical research.

[0003] In existing methods for identifying tectonic environments based on apatite composition, only static geochemical data from a single point on the mineral surface are extracted for classification calculations. However, apatite in actual geological environments often undergoes an early magmatic crystallization stage followed by a later hydrothermal superposition. Conventional single-point static tests are insufficient to reflect the compositional gradient changes from the core to the edge of the mineral, resulting in a lack of sufficient discrimination basis when the algorithm processes samples with complex growth and evolution histories.

[0004] Meanwhile, most existing data classification algorithms adopt a purely data-driven model, which lacks the constraints of prior geological knowledge in the feature extraction and probability output stages. Due to the neglect of the limitations of real physical and chemical conditions such as magma temperature and oxygen fugacity, conventional models may misjudge between tectonic environment categories with completely opposite physical formation conditions when processing geological data with blurred feature boundaries, and output judgment results that violate objective geological laws.

[0005] To obtain dynamic evolution characteristics within minerals and improve discrimination accuracy, some detection procedures attempt to force comprehensive micro-area line scanning analysis on all geological samples. However, this indiscriminate comprehensive scanning method prolongs the instrument's operating time, resulting in high testing costs. For conventional samples with sufficiently clear static geochemical characteristics, line scanning not only generates a large amount of redundant data but also seriously delays the overall analysis progress. Existing technologies struggle to achieve a reasonable balance between instrument testing efficiency and the accuracy of tectonic environment discrimination. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for identifying tectonic environments based on apatite geochemical characteristics. This solves the problems of misjudgment caused by the lack of geological constraints in existing technologies, and low testing efficiency due to indiscriminate full-area scanning.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] The first aspect of this invention provides a method for identifying tectonic environments based on apatite geochemical characteristics, the method comprising the following steps:

[0009] Obtain a test sample target containing baseline particles and evolutionary reserved particles, and save the coordinates and test status parameters of the baseline particles and the evolutionary reserved particles in the database. The initial test status parameters are set to undamaged.

[0010] In-situ micro-regional major element analysis is performed on the reference group particles to calculate the major element characteristic index of the reference group particles to generate a test sequence file. Based on the test sequence file and the coordinates, in-situ trace element analysis is performed on the reference group particles on the test sample target to obtain geochemical composition data, and the test status parameters of the reference group particles are updated to "damaged".

[0011] The geochemical composition data is subjected to outlier screening, detection limit smoothing, and logarithmic ratio transformation to construct the static feature vector of the sample to be tested.

[0012] The static feature vector is input into the classification model, which outputs the original posterior probability of belonging to various tectonic environments. The loss function of the classification model includes a geological origin penalty matrix.

[0013] Calculate the difference between the two largest class probabilities in the original posterior probability. When the difference is less than the confidence threshold, obtain the coordinates of the undamaged evolutionary reserve group particles in the database. Perform a line scan test on the evolutionary reserve group particles based on the obtained coordinates to extract the spatial component evolution gradient features. Combine the static feature vector and the spatial component evolution gradient features to output the final construction environment determination result.

[0014] The above-mentioned technical solution of the present invention sets up a benchmark group and an evolutionary reserve group, and uses static features to make judgments in the initial analysis stage. When the probability difference of the static feature judgment results is less than a set threshold, a line scan test is triggered for the particles of the evolutionary reserve group to supplement the spatial composition evolution data. In addition, by introducing a geological genesis penalty matrix into the loss function of the classification model, the model is constrained by magma temperature and oxygen fugacity conditions, thereby reducing the misjudgment rate of the model output under complex environments.

[0015] Preferably, the reference group particles are single mineral particles with complete crystal morphology, uniform optical characteristics, and no obvious internal zoning structure; the evolution reserved group particles are single mineral particles with a long axis physical length greater than 100 micrometers and visible growth zoning or core-edge structure; the reference group particles and the evolution reserved group particles are arranged alternately and fixed in the same mold to form the sample target to be tested. In this embodiment, the target is classified according to particle morphology and internal optical characteristics. The reference group is used to obtain initial magma composition data, and the evolution reserved group is used to retain the growth zoning structure to extract later fluid superposition information.

[0016] Furthermore, the calculation of the principal element characteristic indices of the benchmark particles to generate a test sequence file includes:

[0017] The mass fractions of chlorine and iron in the reference group particles obtained from the in-situ micro-region major element analysis are extracted to construct a trigger index representing the volatile fraction phase environment; the trigger index is numerically compared with a pre-set physicochemical boundary threshold; if the trigger index is greater than the physicochemical boundary threshold, a fluid-sensitive mode test sequence file is generated, and the single-point integration time of strontium and uranium is extended;

[0018] If the trigger index is less than or equal to the physicochemical boundary threshold, a crystal fractionation mode test sequence file is generated, and the single-point integration time of rare earth elements and strontium elements is extended respectively. In this embodiment, the compositional characteristics of geological samples are evaluated in advance based on the calculation results of major elements, and the accuracy of trace element data acquisition is improved by adjusting the instrument integration time of relevant indicator elements.

[0019] Preferably, the step of performing outlier screening, detection limit smoothing, and logarithmic ratio transformation on the geochemical composition data to construct the static feature vector of the sample to be tested includes:

[0020] The median absolute deviation of the trace element concentration sequence of the same element is calculated for the geochemical composition data. The robust standard deviation fraction is calculated based on the median absolute deviation to screen outliers. For data points below the instrument detection limit after outlier screening, a smoothing substitution operation is performed. The specific element concentration values ​​after smoothing substitution are extracted and the logarithm of the ratio is calculated to map the data from the fixed-constraint space to Euclidean space, and to obtain orthogonal dimension parameters reflecting the volatile matter metasomatism of the magma source region and orthogonal dimension parameters reflecting the degree of deep magma retention, separation and crystallization.

[0021] The acquired traditional scalar features are combined with the orthogonal dimension parameters through mathematical dimension splicing to construct the static feature vector. In this embodiment, logarithmic ratio transformation is used to handle the closure problem caused by the summation effect of component data, and different geological evolution processes are independently characterized by extracting orthogonal dimension parameters.

[0022] Furthermore, the classification model is a deep feedforward neural network, and the geological origin penalty matrix is ​​embedded in the bottom layer of the classification algorithm to construct an asymmetric penalty loss function. The construction principle of the geological origin penalty matrix is ​​as follows: when two tectonic environments are relatively similar in terms of magma temperature and oxygen fugacity physical conditions, a low penalty weight coefficient is assigned; when there are extreme contradictions in the physical formation conditions, a high penalty weight coefficient is assigned. In the process of outputting the original posterior probability, a temperature scaling algorithm is used to introduce a global temperature scalar parameter in the calculation of the normalized exponential function to smoothly adjust the final probability distribution shape of the output layer. In this embodiment, geological and physical conditions are introduced as asymmetric weights into the loss function calculation at the bottom layer of the algorithm to control the weight allocation of the model between the different categories of physical and chemical environments.

[0023] Preferably, the step of performing a line scan test on the evolutionary reserved group particles based on the obtained coordinates to extract spatial composition evolution gradient features includes:

[0024] Real-time images of the target evolutionary reserved group particles are acquired, and the major axis vector direction and physical boundary contour of the target evolutionary reserved group particles are extracted by combining image edge detection algorithms. Using the corresponding coordinates retrieved from the database as the starting point, the endpoint coordinates that touch the physical boundary contour are automatically calculated along the major axis vector direction, thereby constructing a precise line scan trajectory path. Based on the line scan trajectory path, a continuous line scan ablation test extending from the mineral crystal core to the edge is performed on the surface of the target evolutionary reserved group particles to obtain spatial sequence ablation data.

[0025] Based on the spatial sequence erosion data, the differential ratio of the logarithmic ratio of strontium and yttrium concentrations representing the post-magmatic hydrothermal superposition is calculated as a function of the actual physical distance of the line scan. This is used as a feature of the spatial composition evolution gradient. In this embodiment, the test path is obtained by combining image recognition with coordinates, and the degree of fluid metasomatism is characterized by the spatial differential rate of the logarithmic ratio of the characteristic element concentrations.

[0026] Furthermore, the step of combining the static feature vector and the spatial component evolution gradient feature to output the final construction environment determination result includes:

[0027] The static feature vector and the spatial component evolution gradient feature values ​​are tensor-concatenated and fused to construct a fused extended feature vector; the fused extended feature vector is input into a quadratic classification network model in the training convergence state for forward mapping calculation; based on the final identification probability distribution output by the quadratic classification network model, the category name with the largest value in the final identification probability distribution is selected as the final construction environment judgment result with the highest confidence. In this embodiment, the final classification judgment result is obtained by fusing static component sequences and spatial dynamic features.

[0028] Preferably, after calculating the difference between the two largest class probabilities in the original posterior probability, the method further includes:

[0029] When the difference is greater than or equal to the confidence threshold, the construction environment name corresponding to the first major category probability after descending order is directly extracted as the final identification result and the analysis process is terminated. In this embodiment, when the static feature determination meets the confidence requirement, the test step of the evolutionary reserved group particles is omitted, thereby improving the overall detection efficiency.

[0030] A second aspect of this invention provides a system for identifying tectonic environments based on apatite geochemical characteristics, the system comprising:

[0031] The sample preparation module is used to acquire a test sample target containing a reference group of particles and an evolutionary reserved group of particles, and is equipped with a relational database that stores the coordinates and test state parameters of the reference group particles and the evolutionary reserved group particles.

[0032] The component analysis module is equipped with a data flow bus interface, which is used to perform in-situ micro-region major element analysis on the reference group particles to calculate the major element characteristic index, generate a test sequence file based on the major element characteristic index, and perform in-situ trace element analysis based on the test sequence file and the coordinates in the relational database to obtain geochemical composition data.

[0033] The data preprocessing module, connected to the component analysis module, is used to receive the geochemical component data and perform outlier screening, detection limit smoothing, and logarithmic ratio transformation operations.

[0034] The feature extraction module, connected to the data preprocessing module, is used to construct static feature vectors based on the processed data;

[0035] The discrimination module, connected to the feature extraction module, includes a classification model for receiving the static feature vector and outputting the original posterior probability. It is configured to calculate the difference between the two largest class probabilities in the original posterior probability to trigger a spatial tracing instruction for the evolutionary reserved group particles to extract spatial component evolution gradient features, and to output the final construction environment determination result by combining the static feature vector and the spatial component evolution gradient features.

[0036] The system of this invention corresponds to the aforementioned discrimination method. By setting discrimination nodes that connect each module, it automatically issues spatial tracing instructions when the data confidence level is low, driving the component analysis equipment to perform supplementary testing, thereby realizing the control and linkage of the classification algorithm on the hardware testing behavior.

[0037] Preferably, the discrimination module includes a feedback control interface; the feedback control interface is connected to the relational database of the sample preparation module and the component analysis module respectively; the feedback control interface is configured to: when the difference between the probabilities of the two largest categories is less than a preset confidence threshold, trigger the spatial tracing instruction to query the relational database to obtain the coordinates of the undamaged evolutionary reserve group particles with test status parameters, and send the coordinates to the component analysis module to execute a secondary test instruction for continuous line scan ablation test so that the feature extraction module can extract the spatial component evolution gradient features. In this embodiment, the feedback control interface calls the database and the analysis module to systematically complete the supplementary analysis and data integration discrimination of the evolutionary reserve group particles.

[0038] This invention provides a method and system for identifying tectonic environments based on apatite geochemical characteristics. It has the following beneficial effects:

[0039] 1. This invention balances the testing efficiency of geological samples with the accuracy of environmental discrimination. By setting up a baseline group and an evolutionary reserve group during the target preparation stage, and introducing a confidence judgment mechanism based on probability difference in the analysis process, the system can directly output the judgment result and terminate the process after static testing of samples with typical characteristics in the baseline group. Only when the probability difference between the two largest categories is less than the confidence threshold is the line scan test of particles in the evolutionary reserve group triggered to supplement spatial evolution characteristics. This method avoids performing time-consuming full micro-area line scans on all samples, thus controlling instrument time and cost.

[0040] 2. This invention improves the accuracy and reliability of key micro-area geochemical data acquisition. By pre-extracting the mass fractions of chlorine and iron in the baseline particles to construct a trigger index, the genetic environment tendency of the sample is pre-assessed, and different test sequence files are generated accordingly. In the in-situ trace element analysis stage, the instrument integration time of specific environmental indicator elements, such as strontium and uranium in fluid-sensitive mode, or rare earth elements in fractionation mode, is extended in a targeted manner. This solves the problem of insufficient signal-to-noise ratio of low-content characteristic elements in conventional fixed mode, and provides a high-quality data foundation for subsequent feature extraction.

[0041] 3. This invention reduces the misjudgment rate of classification algorithm models when processing complex geological data, which deviates from physical reality. By transforming physicochemical conditions such as magma temperature and oxygen fugacity into a geological origin penalty matrix and embedding it into the asymmetric penalty loss function of the classification model, geological prior rules are imposed on the bottom layer of the model. For tectonic environment categories with extremely contradictory physical formation conditions, the model will generate a high penalty when misjudging, thereby guiding the optimization direction of the data-driven model to conform to objective geological laws and improving the rationality of the final identification probability distribution. Attached Figure Description

[0042] Figure 1This is a system architecture diagram of the present invention;

[0043] Figure 2 This is a flowchart of the method of the present invention;

[0044] Figure 3 This is a scatter plot showing the spatial distribution of particles on the sample target surface and the mapping of database status in this invention.

[0045] Figure 4 This is the timing diagram for the adaptive feedforward control of the present invention;

[0046] Figure 5 This is the logarithmic ratio transform Euclidean space data dimensionality reduction mapping diagram of the present invention;

[0047] Figure 6 This is a heatmap of the geological genesis asymmetric penalty matrix of the present invention;

[0048] Figure 7 This invention provides an evolutionary gradient map of the micro-region composition of the reserved spatial line scan group.

[0049] Figure 8 This is a scatter plot of the three-dimensional feature space distribution of the present invention;

[0050] Figure 9 This is a heatmap of the confusion matrix of the present invention.

[0051] The module includes: 10. Sample preparation module; 20. Component analysis module; 30. Data preprocessing module; 40. Feature extraction module; and 50. Discrimination module. Detailed Implementation

[0052] The technical solutions in 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.

[0053] See attached document Figure 1 The present invention provides a system for identifying tectonic environments based on apatite geochemical characteristics. The system includes: a sample preparation module 10, a component analysis module 20, a data preprocessing module 30, a feature extraction module 40, and a discrimination module 50.

[0054] The sample preparation module 10 includes a crushing device, a mineral separation device, a microscope, and a target preparation tool. The sample preparation module 10 is equipped with a relational database to record the target surface physical coordinates and test status of the apatite particles.

[0055] The component analysis module 20 includes an electron probe analyzer unit and a laser ablation inductively coupled plasma mass spectrometer unit. The component analysis module 20 is equipped with a data transfer bus interface, which connects the electron probe analyzer unit and the laser ablation inductively coupled plasma mass spectrometer unit. The data transfer bus interface reads the major element component data output by the electron probe analyzer unit and generates a control sequence file for the integration time of the corresponding laser ablation inductively coupled plasma mass spectrometer unit.

[0056] The data preprocessing module 30 is connected to the component analysis module 20. The data preprocessing module 30 performs outlier identification and detection limit value replacement of the measurement data, and performs logarithmic ratio transformation on the original component concentration data.

[0057] The feature extraction module 40 is connected to the data preprocessing module 30. The feature extraction module 40 calculates the static geochemical feature vector of a single point and calculates the composition evolution gradient features based on the received spatial sequence test data.

[0058] The discrimination module 50 is connected to the feature extraction module 40. The discrimination module 50 has a built-in classification model containing an asymmetric penalty loss function, which outputs the predicted probability of each constructed environment category. The discrimination module 50 includes a feedback control interface, which is connected to the database of the sample preparation module 10 and the component analysis module 20. When the difference in predicted probability is lower than a set threshold, the feedback control interface outputs a secondary test command for particles at specific coordinates.

[0059] See attached document Figure 2 This invention provides a method for identifying tectonic environments based on apatite geochemical characteristics, the method comprising the following steps:

[0060] S1. Obtain rock samples and extract apatite single mineral particles. Manually select the baseline group particles and the evolutionary reserved group particles. Arrange and fix the baseline group particles and the evolutionary reserved group particles in the same mold to make the sample target to be tested. Establish an initial record in the database and save the coordinates and test status parameters of the baseline group particles and the evolutionary reserved group particles. The initial test status parameters are set to undamaged.

[0061] S2, the electron probe analyzer unit performs in-situ micro-region major element analysis on the baseline group particles and the evolutionary reserved group particles on the test sample target; calculates the major element characteristic index of the baseline group particles, generates a test sequence file based on the major element characteristic index, and adjusts the integration time of different trace elements in the laser ablation inductively coupled plasma mass spectrometer unit; the laser ablation inductively coupled plasma mass spectrometer unit performs in-situ trace element analysis only on the baseline group particles on the test sample target according to the test sequence file and the coordinate data in the database, and updates the test status parameters of the baseline group particles to damaged after the test.

[0062] S3, outlier screening and detection limit smoothing are performed on the acquired geochemical composition data, and logarithmic ratio transformation is performed on the processed data to map the data from the fixed-constraint space to the Euclidean space. The rare earth element distribution characteristics and logarithmic ratio characteristic values ​​are calculated to construct the static feature vector of the sample to be tested.

[0063] S4. Input the static feature vector into the classification model. The classification model outputs the original posterior probability of the sample belonging to each type of tectonic environment. The loss function of the classification model includes the geological origin penalty matrix. Calculate the difference between the probabilities of the two largest categories and determine whether the difference is less than the confidence threshold.

[0064] S5, when the difference is greater than or equal to the confidence threshold, the tectonic environment classification result is directly output; when the difference is less than the confidence threshold, the feedback control interface queries the database to obtain the coordinates of the undamaged evolutionary reserve group particles with test state parameters. The laser ablation inductively coupled plasma mass spectrometer unit performs a line scan test from the center to the edge of the evolutionary reserve group particles according to the coordinates, extracts the spatial composition evolution gradient features corresponding to the line scan test data, and outputs the final tectonic environment determination result by combining the static feature vector and the spatial composition evolution gradient features.

[0065] See attached document Figure 3 In the method for identifying tectonic environments based on the geochemical characteristics of apatite, the physical pretreatment and data archiving of the samples are performed by the sample preparation module 10. The essential physical principle of the above processing operation is to separate and obtain single mineral particles with different crystal states, and to establish a data mapping relationship between the physical three-dimensional space and the test damage attributes in a relational database, thereby providing underlying data support for subsequent automated resource scheduling to avoid physical paradoxes.

[0066] The specific implementation of S1 can be processed through the following sub-steps:

[0067] S101, Obtain rock samples and extract apatite single mineral particles. For the crushing and mineral separation of rock samples, the operator can process them according to conventional mechanical crushing and heavy liquid magnetic separation methods. That is, the rock samples collected in the field are crushed and cut into a particle size range of 0.25 mm to 0.5 mm by a jaw crusher. Then, the crushed rock samples are subjected to continuous gravity and magnetic sieving using a tribromomethane heavy liquid with a density of 2.8 g / cm³ to 3.0 g / cm³ and a magnetic separator with a magnetic field strength of 0.5 amperes to 1.0 amperes. Finally, the coarse mineral bags enriched with apatite minerals are separated.

[0068] S102, the baseline group particles and the evolutionary reserved group particles are manually selected and arranged and fixed in the same mold to make the sample target to be tested. The specific operation procedure is to selectively separate two types of particle sets with different physical morphological characteristics from the mineral coarse package of the same sample under a stereomicroscope. The first type of particle set is the baseline group particles, which are used to represent the primary basic geochemical signal of apatite mineral. The selection criteria are complete crystal morphology, uniform optical characteristics and no obvious internal ring structure. The second type of particle set is the evolutionary reserved group particles, which serve as a backup analysis resource for dealing with the ambiguity in the later category determination.

[0069] The selection criteria for the evolutionary reserve group particles are that the physical length of the long axis is greater than 100 micrometers and there are visible growth rings or core-edge structures. The operator arranges the selected reference group particles and evolutionary reserve group particles at intervals at the bottom of the epoxy resin mold. A physical distance of at least 50 micrometers is strictly maintained between the reference group particles and the evolutionary reserve group particles to prevent plasma cross-contamination or thermal pulse effects from subsequent laser ablation tests. Then, the curing agent is injected into the epoxy resin mold and vacuum degassing and static curing are completed. The cured resin block is then subjected to surface mechanical grinding and polishing treatment so that the core profiles of the reference group particles and the evolutionary reserve group particles are completely flush with the surface of the prepared test sample target.

[0070] S103, measure the coordinates of each particle on the target sample and establish an initial record in the database, save the coordinates and test status parameters of the baseline group particles and the evolution reserved group particles, place the target sample under the microscopic measurement device, establish a global two-dimensional Cartesian coordinate system with a specific reference mark on the edge of the target sample as the origin, the sample preparation module 10 is configured with a relational database to store structured sample metadata, the system accurately determines the two-dimensional physical coordinates of the center point of each exposed particle on the surface of the target sample in the global two-dimensional Cartesian coordinate system, and creates an independent record entry for each apatite particle in the relational database.

[0071] The aforementioned record entries are mathematically and logically represented by metadata vectors:

[0072] ;

[0073] In the formula, This is the metadata vector for apatite particles; Assign a sample identification number; Group classification tags; This represents the horizontal axis coordinate of a specific particle in the global two-dimensional Cartesian coordinate system. This represents the vertical coordinate of a specific particle in the global two-dimensional Cartesian coordinate system. These are the test status parameters.

[0074] The group classification label is used to clarify whether the entity corresponding to a specific coordinate belongs to the baseline group particle or the evolutionary reserved group particle. The test status parameter is used to mark whether the entity particle has suffered irreversible physical morphological damage due to destructive analysis testing. The value range of the test status parameter is strictly set to the values ​​0 and 1. During the system data initialization stage, since all particles on the target sample have not undergone destructive laser ablation analysis testing, the system automatically assigns the test status parameter of all particles in the relational database to the value 0 to represent an undamaged state. Through the aforementioned recording method based on two-dimensional physical coordinate binding, a precise mapping relationship is established between the entity particle array of the physical target surface and the database status parameter. This enables the system to accurately perform spatial addressing based on the value of the test status parameter and automatically avoid damaged particle entities with a test status parameter of value 1 when issuing hardware scheduling instructions.

[0075] See attached document Figure 4 In the method for identifying tectonic environments based on the geochemical characteristics of apatite, the dynamic control of major element acquisition and trace element testing parameters is executed by the composition analysis module 20. The physical principle of the above control logic is to use the small fluctuations of mineral major elements as a priori indication signal of the volatile evolution environment of the magma source region. The priori indication signal is converted into an adjustment instruction for the integration time of a specific trace element through the cross-device data bus interface, thereby changing the hardware resource allocation ratio in the subsequent destructive spectral analysis stage.

[0076] The specific implementation of S2 can be carried out through the following sub-steps:

[0077] S201, the electron probe analyzer unit performs in-situ micro-region major element analysis on the reference group particles and the evolved reserved group particles on the test sample target. For the setting of electron probe test parameters and the acquisition of surface excitation data, the operator can process them according to the existing micro-region in-situ composition testing methods, that is, configure the accelerating voltage of the electron probe analyzer unit to be 15 kV to 20 kV and the electron beam current to be 10 nanoamps to 20 nanoamps. The operator controls the electron beam to perform non-destructive energy excitation operation on the surface of the reference group particles and the evolved reserved group particles of the test sample target, and collects X-ray energy spectrum data of specific components such as calcium and iron inside each particle to calculate the corresponding mass fraction.

[0078] S202, the component analysis module 20 calculates the major element characteristic indices of the reference group particles and generates test sequence files to adjust the integration time of specific trace elements in the laser ablation inductively coupled plasma mass spectrometer unit. The data transfer bus interface configured in the component analysis module 20 extracts the mass fractions of chlorine and iron in the reference group particles in real time. The trigger index representing the volatile fraction phase environment is constructed through the system's built-in calculation program. The calculation logic of the trigger index is shown in the following formula:

[0079] ;

[0080] In the formula: For triggering the index; The mass fraction of chlorine in the particles of the baseline group; This represents the mass fraction of iron in the particles of the baseline group.

[0081] The component analysis module 20 compares the calculated trigger index with the pre-set physicochemical boundary threshold. The physicochemical boundary threshold is obtained by statistically calibrating the trigger index of standard samples of known tectonic environments based on empirical data. The range of the physicochemical boundary threshold is set to -1.5 to -1.0. When the trigger index is greater than the physicochemical boundary threshold, the system determines that the test object originates from a convergent edge fluid environment rich in volatiles. The data transfer bus interface automatically generates a fluid-sensitive mode test sequence file accordingly. In the fluid-sensitive mode test sequence file, the data transfer bus interface extends the single-point integration time of strontium and uranium elements from the conventionally set 10 milliseconds to 20 milliseconds.

[0082] When the trigger index is less than or equal to the physicochemical boundary threshold, the system determines that the test object originates from the intraplate crystallization differentiation-dominant environment. Based on this, the data transfer bus interface automatically generates a crystal fractionation mode test sequence file. In the crystal fractionation mode test sequence file, the data transfer bus interface extends the single-point integration time of rare earth elements and strontium elements from the conventionally set 10 milliseconds to 20 milliseconds.

[0083] S203, the laser ablation inductively coupled plasma mass spectrometer unit performs in-situ trace element analysis on the reference group particles on the target sample based on the test sequence file and coordinate data in the database. After the test, it updates the test status parameters of the reference group particles to "damaged". The laser ablation inductively coupled plasma mass spectrometer unit loads the test sequence file generated by the data transfer bus interface through the internal communication interface. The system reads the group classification labels stored in the relational database and controls the laser emitter to strictly avoid the coordinate positions of the particles in the evolution reserved group according to the group classification labels.

[0084] The laser emitter generates erosion pits with a diameter of 30 to 40 micrometers on the surface of the reference group particles based solely on their two-dimensional physical coordinates to extract trace element mass concentration data. After the laser ablation process for a specific reference group particle ends and the mass spectrometer completes the trace data acquisition operation, the component analysis module 20 sends a state overwrite instruction to the relational database of the sample preparation module 10. The relational database receives the state overwrite instruction and forcibly rewrites the test state parameter of the corresponding reference group particle from the value 0 to the value 1. The test state rewrite mechanism ensures that the reference group particles destroyed by the high-energy laser are completely logically shielded in the subsequent system feedback scheduling stage.

[0085] See attached document Figure 5 In the method for identifying tectonic environments based on apatite geochemical characteristics, the preprocessing and feature construction of test data are jointly performed by the data preprocessing module 30 and the feature extraction module 40. The underlying principle of the above processing mechanism is that geochemical composition data is typical composition data and has fixed-constraint characteristics. Direct statistical analysis will produce spurious correlation problems. It is necessary to use specific mathematical transformations to map the original concentration data from the closed simplex domain to the completely open Euclidean space, thereby ensuring the mathematical unbiasedness of the subsequently extracted geological feature parameters.

[0086] The specific implementation of S3 can be carried out through the following sub-steps:

[0087] S301, the data preprocessing module 30 receives the geochemical composition data transmitted by the composition analysis module 20 and performs outlier screening and detection limit smoothing on the geochemical composition data. The data preprocessing module 30 calculates the median absolute deviation for the same trace element concentration sequence within the same sample. The calculation logic for the median absolute deviation of the element concentration sequence is shown in the following formula:

[0088] ;

[0089] In the formula, This represents the absolute deviation of the median of the elemental concentration sequence. The first in the trace element concentration sequence of the same element One test value; This is a sequence of trace element concentrations of the same element; The sign of the median function is calculated; The sequence index number of the test value.

[0090] The data preprocessing module 30 calculates the robust standard deviation of each test value in the trace element concentration sequence of the same element based on the median absolute deviation of the element concentration sequence. The calculation logic of the robust standard deviation is shown in the following formula:

[0091] ;

[0092] In the formula, For the first Robust standard deviation score of each test value; The first in the trace element concentration sequence of the same element One test value; This represents the absolute deviation of the median of the elemental concentration sequence. This is a sequence of trace element concentrations of the same element; The sign of the median function is calculated; The sequence index number of the test value.

[0093] Data preprocessing module 30 will... The absolute value of the robust standard deviation of each test value is compared with the outlier detection threshold, which is set to a constant of 3 based on the statistical Laida criterion. When the absolute value of the robust standard deviation is greater than the constant of 3, the data preprocessing module 30 marks the corresponding test value as an outlier and removes it from the element concentration sequence. For data points where the test value is lower than the instrument's detection limit, the data preprocessing module 30 performs a smoothing substitution operation. The smoothing substitution rule is to replace the original test value with the sum of 0.5 times the instrument's detection limit and the anti-divergence constant. The calculation logic of the smoothing substitution is shown in the following formula:

[0094] ;

[0095] In the formula, The test value after smooth substitution; For the first The instrument detection limit value corresponding to each test value; The sequence index number of the test value; To prevent divergence constants.

[0096] The anti-divergence constant is used to prevent mathematical calculation crashes caused by the denominator being zero or negative infinity during subsequent logarithmic operations. The value range of the anti-divergence constant is set to 0.000001 to 0.0001. In this embodiment, the anti-divergence constant is preferably determined to be 0.00001.

[0097] S302, the data preprocessing module 30 performs a logarithmic ratio transformation on the processed data to map the data from a fixed-constraint space to Euclidean space. The data preprocessing module 30 extracts the specific element concentration values ​​after smoothing and substitution, performs a logarithmic ratio operation, and calculates the orthogonal dimension parameters reflecting the volatile matter alteration in the magma source region and the orthogonal dimension parameters reflecting the degree of deep magma retention, separation, and crystallization. By performing a logarithmic ratio operation, the inherent collinearity problem between the original element variables is reduced. The calculation logic of the orthogonal dimension parameters reflecting the volatile matter alteration in the magma source region is shown in the following formula:

[0098] ;

[0099] In the formula, To reflect the orthogonal dimensional parameters of volatile matter metasomatism in the magma source region; The chlorine concentration value after smoothing the substitution; This represents the iron concentration value after smoothing the substitution.

[0100] The data preprocessing module 30 extracts the concentration ratio of cerium to strontium based on the difference in the partition coefficients of rare earth elements and alkaline earth metals during the evolution of magmas from the same source. The calculation logic of the orthogonal dimension parameter reflecting the degree of deep magma retention, separation, and crystallization is shown in the following formula:

[0101] ;

[0102] In the formula, An orthogonal dimensional parameter reflecting the degree of deep magma retention, separation, and crystallization; The cerium concentration value after smoothing substitution; This represents the strontium concentration value after smoothing substitution.

[0103] S303, the feature extraction module 40 calculates the rare earth element distribution characteristics and logarithmic ratio characteristic values ​​based on the data mapped to Euclidean space to construct the static feature vector of the sample to be tested. In addition to receiving orthogonal dimension parameters, the feature extraction module 40 also performs standardized division operations on the concentration of specific rare earth elements according to the geochemical standard chondrite values. The feature extraction module 40 obtains traditional scalar features representing the origin, such as europium anomaly index and cerium anomaly index, through standardized calculation. The europium anomaly index is defined as the geometric mean of the europium element concentration after chondrite standardization divided by the standardized samarium and gadolinium element concentrations. The cerium anomaly index is defined as the geometric mean of the cerium element concentration after chondrite standardization divided by the standardized lanthanum and praseodymium element concentrations.

[0104] The feature extraction module 40 performs a mathematical dimension concatenation operation on the calculated traditional scalar features and orthogonal dimension parameters. The static feature vector is constructed by the system's underlying layer into a one-dimensional floating-point tensor containing 10 to 15 feature dimensions. The specific order of the feature dimensions is europium anomaly index, cerium anomaly index, and the aforementioned... and In order to meet the preset tensor dimension and fully cover magma evolution information, the multiple sets of log-ratio orthogonal dimension parameters also include three conventional trace element genetic correlation features: the log-ratio of thorium to uranium concentration, the log-ratio of strontium to yttrium concentration, and the log-ratio of lanthanum to ytterbium concentration. Finally, the feature extraction module 40 outputs a static feature vector with a fixed tensor dimension format. The fixed-format static feature vector is directly passed to the discrimination module 50 as a standardized input parameter for subsequent neural network pattern recognition operations.

[0105] See attached document Figure 6 In the method for identifying tectonic environments based on apatite geochemical characteristics, probability prediction and uncertainty assessment are performed by the discrimination module 50. The underlying physical logic of the above processing mechanism is to map nonlinear geochemical element combination relationships through a multi-layer neural network and combine them with geological prior knowledge to constrain the classification boundary, thereby outputting a tectonic environment probability distribution with physical genetic significance.

[0106] The specific implementation of S4 can be carried out through the following sub-steps:

[0107] S401, the discrimination module 50 internally constructs a deep feedforward neural network as a classification model to perform pattern recognition tasks. The deep feedforward neural network sequentially includes an input layer, two fully connected hidden layers, and an output layer. The input layer directly receives static feature vectors in one-dimensional floating-point tensor format with a dimension of 10 to 15 from the feature extraction module 40. The number of neurons in the first fully connected hidden layer is set to 64 to 128, and the number of neurons in the second fully connected hidden layer is set to 32 to 64. A random deactivation network layer with a dropout rate parameter of 0.2 to 0.3 is configured between the two fully connected hidden layers to prevent overfitting during model training. All fully connected hidden layers use modified linear units as activation functions to capture the nonlinear implicit correlations between geochemical variables. The number of neurons in the output layer is 4, which is consistent with the preset total number of tectonic environment categories.

[0108] The output layer uses a normalized exponential function to map the output signal into a probability distribution vector with a sum of 1. Each element in the probability distribution vector directly corresponds to the operational probability meaning of the sample belonging to one of the four basic geological tectonic environments: mid-ocean ridge, island arc, ocean island, and continental margin. The training samples of the classification model are derived from real in-situ test data of apatite with clear tectonic environment attributes identified in the global standard rock geochemical database. The labels of the training samples are marked using a one-hot encoding format to meet the underlying data flow requirements of the neural network cross-entropy calculation.

[0109] S402, the discrimination module 50 defines the misclassification penalty weights based on the differences in the physicochemical properties of magma source regions between geodynamic environments and constructs a geological origin penalty matrix. The construction principle of the geological origin penalty matrix is ​​to assign a low penalty weight coefficient of 1.0 to 1.5 when the physical conditions of magma temperature and oxygen fugacity of two tectonic environments are relatively similar, and to assign a high penalty weight coefficient of 3.0 to 5.0 when the physical formation conditions of two tectonic environments are extremely contradictory. The discrimination module 50 embeds the geological origin penalty matrix into the bottom layer of the classification algorithm to construct an asymmetric penalty loss function to force the model to train and converge in a direction that conforms to geological laws.

[0110] The calculation logic of the asymmetric penalty loss function is shown in the following formula:

[0111] ;

[0112] In the formula, The calculated value is for the asymmetric penalty loss function; This represents the total number of preset construction environment categories; To construct the index number of the environment category; For the true category And the predicted category is The corresponding geological genesis penalty matrix element values ​​at that time; Construct a real-world environment category label index for the training samples; One-hot encoding of the true category label in the first... The value at each position; The first prediction output of the classification model The original posterior probability of the class-constructed environment; The preset penalty balance coefficient is set to a value ranging from 0.1 to 1.0. For classification Not equal to the true category When the value of this item is 1, the misjudgment penalty weight in the penalty matrix is ​​activated.

[0113] The discrimination module 50 uses an adaptive moment estimation optimization algorithm to minimize the aforementioned asymmetric penalty loss function during the model training phase, and repeatedly updates the weight parameters of the deep feedforward neural network until the classification model reaches convergence. The batch size during the model training phase is set to 32 to 128, and the initial learning rate is set to 0.001 to 0.005. The specific criteria for determining convergence are that the asymmetric penalty loss function decreases by less than 0.001 for 10 consecutive iterations or the total number of training iterations reaches the preset 200 to 500 times.

[0114] S403, the discrimination module 50 directly inputs the static feature vector generated in the data preprocessing stage as fixed-dimensional input data into the trained classification model. The classification model calculates and outputs the original posterior probability of the test sample belonging to various construction environments through forward propagation at the internal neuron node level. To avoid the drawback of overconfident prediction in neural networks and to ensure that the output probability truly reflects the uncertainty of the prediction, the discrimination module 50 uses a temperature scaling algorithm to perform numerical calibration on the original posterior probability. The temperature scaling algorithm introduces a global temperature scalar parameter with a value of 1.5 to 2.5 in the calculation of the normalized exponential function to smoothly adjust the final probability distribution shape of the output layer. The calculation logic of the normalized exponential function with the global temperature scalar parameter is shown in the following formula:

[0115] ;

[0116] In the formula, For the calibrated first The original posterior probability of the class-constructed environment; For the output layer of the classification model before normalization, the corresponding number of... The original logical output value of the class construction environment; This is a global temperature scalar parameter; This represents the total number of preset construction environment categories; This is the index number for the cumulative summation algorithm; For the output layer of the classification model, corresponding to the first... The original logical output value of the class construction environment.

[0117] S404, the discrimination module 50 extracts the probability distribution vector generated by the output layer and sorts all the probability values ​​in the probability distribution vector in descending order. The discrimination module 50 selects the probability values ​​of the first and second largest categories after the descending order and calculates the difference between the two largest category probabilities. The difference between the two largest category probabilities directly reflects the degree of decision hesitation of the classification model between the two most likely geological causes in the physical business logic. The discrimination module 50 compares the calculated difference between the two largest category probabilities with the pre-set confidence threshold.

[0118] The confidence threshold is calculated and calibrated during the cross-validation stage based on the statistical distribution curve of the historical test dataset. The range of the confidence threshold is limited to 0.2 to 0.4. In this embodiment, in order to balance the test cost and the prediction accuracy, the confidence threshold is preferably determined to be 0.3. The discrimination module 50 determines whether to block the normal output process and trigger a cross-boundary physical analysis scheduling instruction to the feedback control interface based on the aforementioned numerical judgment comparison results.

[0119] See attached document Figure 7 In the method for identifying tectonic environments based on apatite geochemical characteristics, hardware physical scheduling and secondary disambiguation discrimination operations are jointly executed by discrimination module 50 and composition analysis module 20. The underlying physical logic of the above closed-loop processing mechanism is that when the static geochemical characteristics of a single grain are insufficient to provide a clear indication of the geological tectonic environment, the system can actively obtain the composition gradient change sequence retained by the crystallization kinetic process within the homologous associated minerals by crossing the physical barrier of the damaged state. By utilizing the dynamic evolution information of composition in the spatial physical dimension, the decision ambiguity range of the static classification model can be broken, thereby outputting the final identification conclusion.

[0120] The specific implementation of S5 can be carried out through the following sub-steps:

[0121] S501, the discrimination module 50 performs a triage decision based on the comparison result of the difference between the two largest class probabilities and the confidence threshold. When the difference between the two largest class probabilities is greater than or equal to the confidence threshold, the system determines that the current static feature has a strong environmental indication significance. The discrimination module 50 directly extracts the name of the tectonic environment corresponding to the first largest class probability after descending order as the final identification result and terminates the analysis process. When the difference between the two largest class probabilities is less than the preset confidence threshold, the system determines that it has encountered a geologically ambiguous area and intercepts the initial judgment result. The discrimination module 50 sends a spatial tracing instruction to the relational database of the sample preparation module 10. After receiving the spatial tracing instruction, the relational database crosses the damaged reference group particles with a test state parameter of value 1 and retrieves the physical coordinates of the evolutionary reserved group particles with a test state parameter of value 0 in the same sample from the database. Through this physical state avoidance mechanism, the adaptive cross-boundary scheduling of limited test resources on the same physical target surface is realized.

[0122] S502, the component analysis module 20 controls the laser ablation device to perform in-situ line scanning operation and extract micro-region evolution gradient features based on the newly retrieved physical coordinates of the evolutionary reserved group particles. The component analysis module 20 sends the physical coordinates of the evolutionary reserved group particles to the laser ablation inductively coupled plasma mass spectrometer unit. The system acquires real-time images of the target evolutionary reserved group particles through the coaxial microscopic imaging unit of the laser ablation device, and extracts the major axis vector direction and physical boundary contour of the particles by combining the image edge detection algorithm. Using the physical coordinates recorded in the database as the starting point, the system automatically calculates the endpoint coordinates of touching the physical boundary contour along the major axis vector direction, thereby constructing a precise line scanning trajectory path.

[0123] The in-situ line scanning motion control of the laser ablation equipment is handled according to the existing spatial stepping test method. The operator, in accordance with the system instructions, sets the laser spot diameter to 10 to 15 micrometers and configures the scanning translation speed to 2 to 5 micrometers per second. The laser emitter performs a continuous line scanning ablation test on the surface of the evolved reserved group of particles, extending from the mineral crystal core to the edge, to obtain spatial sequence ablation data based on the calculated line scanning trajectory path.

[0124] The composition analysis module 20 receives spatial sequence erosion data and extracts the logarithmic ratio of strontium and yttrium concentrations, representing the post-magmatic hydrothermal superposition. The composition analysis module 20 calculates the differential ratio of the aforementioned logarithmic ratio as a function of the actual physical distance of the line scan, as the composition evolution gradient feature. The calculation logic of the micro-region composition evolution gradient feature is shown in the following formula:

[0125] ;

[0126] In the formula, The numerical value represents the gradient characteristic of component evolution; Strontium concentration values ​​are reserved for the edge regions of pre-evolutionary particles; Yttrium concentration values ​​in the edge region of reserved particles for evolution; Strontium concentration values ​​in the core region of reserved particles for evolution; Yttrium concentration values ​​are reserved for the core region of the pre-evolutionary particle group; This represents the actual physical distance from the start point to the end point of the line scan.

[0127] The actual physical distance from the start point to the end point of the line scan is determined by the development volume of the mineral single crystal. The range of the actual physical distance from the start point to the end point of the line scan is limited to between 50 micrometers and 150 micrometers by the hardware security boundary program.

[0128] S503, the discrimination module 50 fuses static and dynamic features and uses a quadratic classification network model for final judgment to output a clear construction environment type. The discrimination module 50 performs tensor concatenation and fusion of the static feature vector generated by the previous feature extraction module 40 and the component evolution gradient feature value to construct a fused extended feature vector. The tensor dimension of the fused extended feature vector is equal to the sum of the dimension of the static feature vector and the value 1. The discrimination module 50 internally constructs a quadratic classification network model of the specific type of shallow multilayer perceptron. The quadratic classification network model consists of an input layer, an intermediate fully connected hidden layer, and a classification output layer with a normalized exponential function. The number of nodes in the input layer is strictly equal to the dimension of the fused extended feature vector. The intermediate fully connected hidden layer is configured with 32 to 64 neuron nodes and uses a modified linear unit to handle nonlinear mapping relationships. The classification output layer contains 4 nodes to output the business classification probability of the test sample belonging to the mid-ocean ridge, island arc, ocean island, and continental margin, respectively.

[0129] The training samples for the secondary classification network model are specifically selected from a complex environment apatite dataset with obvious core-edge-ring structure evolution characteristics, accumulated from the system's historical in-situ line scanning tests and verified and labeled by human geological experts. The input features of the samples are preprocessed multi-dimensional fusion extended feature vectors, and the true labels of the samples are defined as known tectonic environment categories labeled using a one-hot encoding format. The model training steps include a discrimination module 50 that divides the training samples into batches and inputs them sequentially into the secondary classification network model for forward propagation to calculate the predicted probability distribution. The discrimination module 50 uses the standard cross-entropy loss function to quantify the difference error between the predicted probability distribution calculated by forward propagation and the one-hot encoded labels. The system then executes the backpropagation algorithm to calculate the gradient and uses the optimizer to update the network weight parameters until the secondary classification network model reaches the convergence threshold.

[0130] The discrimination module 50 inputs the fused extended feature vector obtained from the field test into the secondary classification network model that is in the training convergence state for forward mapping calculation. The secondary classification network model overcomes the data limitations of local overlapping areas based on dynamically grown historical data and static feature tensors, and outputs the final identification probability distribution. The discrimination module 50 selects the category name with the largest value in the final identification probability distribution as the construction environment type with the highest confidence and outputs it as the final decision conclusion of the system's global discrimination to the human-computer interaction interface. It can be seen that the above-mentioned homologous secondary verification logic can improve the system's environmental discrimination accuracy in complex geological backgrounds.

[0131] To aid in understanding the technical solution of this invention, the following is an application example based on the discrimination of multi-stage magmatic rock target areas in complex collisional orogenic belts.

[0132] In the exploration of a concealed mineral area in a collisional orogenic belt, porphyry samples with complex compositions were collected. This area has historically experienced oceanic crust subduction and subsequent continental collision. The magmatic genesis overlaps between island arc and active continental margin environments. Traditional whole-rock geochemical methods cannot provide clear conclusions. The system of this invention operates automatically in this scenario, and the specific process is provided below.

[0133] The sample preparation module crushes the porphyry sample and sorts out the apatite particles. Under a microscope, the operator selects 30 optically uniform particles without rings as the baseline particles and 15 particles with visible core-edge structures and a long axis of 120 micrometers as the evolution reserve particles. They are then made into the same epoxy resin target and polished. The system records the two-dimensional physical coordinates of these 45 particles in the database and sets all the initial test state parameters to undamaged.

[0134] The electron probe analyzer unit performs non-destructive major element analysis on the baseline and evolutionary reserve particles. After the composition analysis module extracts the data, it calculates that the logarithmic ratio of the chlorine and iron mass fractions of the baseline particles is higher than the physicochemical boundary threshold. Based on this, the system automatically triggers the fluid-sensitive mode. The data transfer bus interface extends the single-point integration time of strontium and uranium in the laser ablation inductively coupled plasma mass spectrometer unit from 10 milliseconds to 20 milliseconds.

[0135] The laser ablation inductively coupled plasma mass spectrometer unit avoids the evolutionary reserved group of particles based on the database coordinates and only performs in-situ ablation tests on 30 reference group particles. After the test, the database updates the state parameters of these 30 reference group particles to damaged. The data preprocessing module then performs outlier screening, detection limit smoothing, and logarithmic ratio transformation on the collected trace element data. Based on this, the feature extraction module constructs a static feature vector that includes europium anomalies, cerium anomalies, and multiple sets of orthogonal dimension parameters.

[0136] The discrimination module inputs the static feature vector into the classification model with a built-in asymmetric penalty loss function. For a test point, the original posterior probability distribution output by the classification model is 2.3% for mid-ocean ridges, 1.1% for ocean islands, 49.2% for island arcs, and 47.4% for continental margins. The discrimination module calculates that the difference between the probability of the first class and the probability of the second class is 0.018. This value is less than the system's preset confidence threshold of 0.30. The system intercepts the initial judgment result and sends a spatial tracing command to the feedback control interface.

[0137] The feedback control interface queries the database and obtains the coordinates of undamaged evolutionary reserved particles as test status parameters. The laser ablation inductively coupled plasma mass spectrometer unit extracts the major axis vector direction of the particles based on these coordinates using an image edge detection algorithm and performs a continuous line scan test with a physical distance of 85 micrometers from the starting point to the ending point. The composition analysis module extracts spatial sequence ablation data and calculates the composition evolution gradient features. The discrimination module concatenates the static feature vector with the composition evolution gradient features to construct a fused extended feature vector and inputs it into the secondary classification network model. The identification probability distribution output by the secondary classification network model is 89.7% for continental margins, 8.1% for island arcs, and 2.2% for others. The system takes the category with the largest value, continental margin, as the final determination result of the tectonic environment type.

[0138] The experiment collected 145 geological body samples from mid-ocean ridges, island arcs, ocean islands, and continental margin tectonic environments, which were calibrated by isotope geochronology and whole-rock geochemistry. Data from 870 apatite test points were extracted, of which 35% of the samples were pre-marked by geological experts as easily confused samples.

[0139] The comparison scheme includes three methods: Method A uses geochemical two-dimensional scatter projection and empirical boundary lines for classification; Method B uses a preprocessing module, a static feature extraction module, and a feedforward neural network to output the highest probability category; and Method C uses threshold interception, database cross-boundary addressing scheduling, and a secondary fusion dynamic feature network mechanism.

[0140] The software generates a scatter plot of the three-dimensional feature space distribution and a heatmap of the confusion matrix. The kernel density contour projection of the scatter plot of the three-dimensional feature space shows that there is point cluster overlap between method A and method B in the boundary region between the island arc and the continental margin. After introducing the evolution gradient features of the micro-region by line scanning, method C causes the overlapping point clusters to be physically separated on the hyperplane.

[0141] Table 1: Comparison of Performance Evaluation Indicators of Different Test Methods for Identifying Apatite Tectonic Environments

[0142]

[0143] According to Table 1 and Appendix Figure 8 It can be seen that the overall accuracy of method C is 95.14%, the overall accuracy of method B is 81.79%, the resolution rate of easily confused fuzzy samples of method C is 89.05%, and the resolution rate of easily confused fuzzy samples of method B is 38.51%. This invention utilizes the compositional evolution gradient characteristics to distinguish static geochemical data with thermodynamic and crystallographic similarities.

[0144] When distinguishing between island arc and continental margin environments, the recall rate of the system in this invention for the island arc environment is 93.68% and the precision rate for the continental margin environment is 94.27%. The system reduces the probability of misclassification by using a classification model that includes an asymmetric penalty loss function. The color of the misclassified blocks in the off-diagonal region of the confusion matrix heatmap becomes lighter. The introduction of a quadratic line scan mechanism increases the average test time per sample to 22.3 minutes. The system relies on the physical coordinates and test state parameters recorded in the relational database to cross the baseline group particles with a test state parameter of value 1 and retrieve and call the evolutionary reserved group particles with a test state parameter of value 0. The physical damage rate of the target particles is 68.2%, retaining the single mineral sample entity and reducing resource consumption, thus completing the hardware control task.

[0145] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying tectonic environments based on apatite geochemical characteristics, characterized in that, Includes the following steps: Obtain a test sample target containing baseline particles and evolutionary reserved particles, and save the coordinates and test status parameters of the baseline particles and the evolutionary reserved particles in the database. The initial test status parameters are set to undamaged. In-situ micro-regional major element analysis is performed on the reference group particles to calculate the major element characteristic index of the reference group particles to generate a test sequence file. Based on the test sequence file and the coordinates, in-situ trace element analysis is performed on the reference group particles on the test sample target to obtain geochemical composition data, and the test status parameters of the reference group particles are updated to "damaged". The geochemical composition data is subjected to outlier screening, detection limit smoothing, and logarithmic ratio transformation to construct the static feature vector of the sample to be tested. The static feature vector is input into the classification model, which outputs the original posterior probability of belonging to various tectonic environments. The loss function of the classification model includes a geological origin penalty matrix. Calculate the difference between the two largest class probabilities in the original posterior probability. When the difference is less than the confidence threshold, obtain the coordinates of the undamaged evolutionary reserve group particles in the database. Perform a line scan test on the evolutionary reserve group particles based on the obtained coordinates to extract the spatial component evolution gradient features. Combine the static feature vector and the spatial component evolution gradient features to output the final construction environment determination result.

2. The method for identifying tectonic environments based on apatite geochemical characteristics according to claim 1, characterized in that, The reference group particles are single mineral particles with complete crystal morphology, uniform optical characteristics, and no obvious internal ring structure; the evolution reserved group particles are single mineral particles with a long axis physical length greater than 100 micrometers and visible growth rings or core-edge structures. The reference group particles and the evolutionary reserved group particles are arranged alternately and fixed in the same mold to form the sample target to be tested.

3. The method for identifying tectonic environments based on apatite geochemical characteristics according to claim 1, characterized in that, The calculation of the major element characteristic indices of the benchmark group particles to generate test sequence files includes: The mass fractions of chlorine and iron in the reference group particles obtained by in-situ micro-region major element analysis were extracted to construct a trigger index representing the volatile fraction phase environment. The triggering index is numerically compared with a pre-set physicochemical boundary threshold; If the triggering index is greater than the physicochemical boundary threshold, a fluid-sensitive mode test sequence file is generated, and the single-point integration time for strontium and uranium is extended; if the triggering index is less than or equal to the physicochemical boundary threshold, a crystal fractionation mode test sequence file is generated, and the single-point integration time for rare earth elements and strontium is extended respectively.

4. The method for identifying tectonic environments based on apatite geochemical characteristics according to claim 1, characterized in that, The process of screening outliers, smoothing detection limits, and performing logarithmic ratio transformation on the geochemical composition data to construct a static feature vector for the sample to be tested includes: The median absolute deviation of the trace element concentration sequence of the same element is calculated from the geochemical composition data, and the robust standard deviation fraction is calculated based on the median absolute deviation for outlier screening. A smoothing substitution operation is performed on data points below the instrument's detection limit after outlier screening. The concentration values ​​of specific elements after smoothing substitution are extracted and the logarithm of the ratio is calculated to map the data from the fixed and constrained space to the Euclidean space, so as to obtain the orthogonal dimension parameters reflecting the volatile matter metasomatism of the magma source area and the orthogonal dimension parameters reflecting the degree of deep magma retention, separation and crystallization. The static feature vector is constructed by concatenating and combining the acquired traditional scalar features with the orthogonal dimension parameters using mathematical dimensions.

5. The method for identifying tectonic environments based on apatite geochemical characteristics according to claim 1, characterized in that, The classification model is a deep feedforward neural network, and the geological genesis penalty matrix is ​​embedded in the bottom layer of the classification algorithm to construct an asymmetric penalty loss function. The principle for constructing the geological genesis penalty matrix is ​​as follows: when two tectonic environments are relatively similar in terms of magma temperature and oxygen fugacity physical conditions, a low penalty weight coefficient is assigned; when there are extreme contradictions in the physical formation conditions, a high penalty weight coefficient is assigned. In the process of outputting the original posterior probability, a temperature scaling algorithm is used to introduce a global temperature scalar parameter in the calculation of the normalized exponential function to smoothly adjust the final probability distribution shape of the output layer.

6. The method for determining tectonic environments based on apatite geochemical characteristics according to claim 1, characterized in that, The step of performing a line scan test on the evolutionary reserved group particles based on the obtained coordinates to extract spatial component evolution gradient features includes: Real-time images of the target evolutionary reserved group particles are acquired, and the major axis vector direction and physical boundary contour of the target evolutionary reserved group particles are extracted by combining image edge detection algorithms. Using the corresponding coordinates retrieved from the database as the starting point, the endpoint coordinates that touch the physical boundary contour are automatically calculated along the direction of the major axis vector, thereby constructing a precise line scan trajectory path; Based on the described line scan trajectory path, a continuous line scan ablation test extending from the mineral crystal core to the edge is performed on the surface of the target evolution reserved group particles to obtain spatial sequence ablation data; Based on the spatial sequence erosion data, the differential ratio of the logarithmic ratio of strontium to yttrium concentrations, representing the post-magmatic hydrothermal superposition, is calculated as a function of the actual physical distance of the line scan, and is used as a feature of the spatial composition evolution gradient.

7. The method for determining tectonic environments based on apatite geochemical characteristics according to claim 1, characterized in that, The final construction environment determination result is output by combining the static feature vector and the spatial component evolution gradient feature, including: The static feature vector is fused with the spatial component evolution gradient feature value by tensor concatenation to construct a fused extended feature vector. The fused extended feature vector is input into a quadratic classification network model that is in the training convergence state for forward mapping calculation; Based on the final identification probability distribution output by the quadratic classification network model, the category name with the largest value in the final identification probability distribution is selected as the final construction environment determination result with the highest confidence.

8. The method for determining tectonic environments based on apatite geochemical characteristics according to claim 1, characterized in that, After calculating the difference between the two largest class probabilities in the original posterior probability, the method further includes: When the difference is greater than or equal to the confidence threshold, the constructed environment name corresponding to the first major category probability after descending order is directly extracted as the final identification result and the analysis process is terminated.

9. A system for determining tectonic environments based on apatite geochemical characteristics, characterized in that, The method for determining tectonic settings based on apatite geochemical characteristics according to any one of claims 1 to 7, the system comprising: The sample preparation module is used to acquire a test sample target containing a reference group of particles and an evolutionary reserved group of particles, and is equipped with a relational database that stores the coordinates and test state parameters of the reference group particles and the evolutionary reserved group particles. The component analysis module is equipped with a data flow bus interface, which is used to perform in-situ micro-region major element analysis on the reference group particles to calculate the major element characteristic index, generate a test sequence file based on the major element characteristic index, and perform in-situ trace element analysis based on the test sequence file and the coordinates in the relational database to obtain geochemical composition data. The data preprocessing module, connected to the component analysis module, is used to receive the geochemical component data and perform outlier screening, detection limit smoothing, and logarithmic ratio transformation operations. The feature extraction module, connected to the data preprocessing module, is used to construct static feature vectors based on the processed data; The discrimination module, connected to the feature extraction module, has a built-in classification model for receiving the static feature vector and outputting the original posterior probability. It is configured to calculate the difference between the two largest class probabilities in the original posterior probability to trigger the spatial tracing instruction for the evolutionary reserved group particles to extract the spatial component evolution gradient features, and combine the static feature vector and the spatial component evolution gradient features to output the final construction environment determination result.

10. The system for determining tectonic environments based on apatite geochemical characteristics according to claim 9, characterized in that, The discrimination module includes a feedback control interface; The feedback control interface is connected to the relational database of the sample preparation module and the component analysis module, respectively. The feedback control interface is configured such that when the difference between the probabilities of the two largest categories is less than a preset confidence threshold, the spatial tracing instruction is triggered to query the relational database to obtain the coordinates of the undamaged evolutionary reserved group particles with test status parameters, and the coordinates are sent to the composition analysis module to execute a secondary test instruction for continuous line scan erosion test so that the feature extraction module can extract the spatial composition evolution gradient features.

Images