102 results about "Spatial aggregation" patented technology

The invention belongs to the technical field of target detection, and discloses an industrial surface defect detection method based on an improved real-time target detection model, and the method comprises the steps: introducing a multi-attention convolution space aggregation module into a feature extraction module of an RTDETR model, enhancing the attention capability of the model to an industrial surface defect region through a space attention mechanism, and improving the detection precision of the industrial surface defect region. Background interference is suppressed, and the defect identification precision is improved; redundant calculation is reduced by using depth separable convolution and a channel attention mechanism, the calculation complexity of the model is reduced, the detection speed is improved, and real-time detection is realized. According to the method, a multi-scale semantic decoding fusion module is used for performing multi-scale feature fusion, and the detection capability on small targets and complex defects is remarkably improved by enhancing information interaction among different scale features. According to the industrial surface defect detection method, surface defect detection in various complex industrial environments can be realized, and the capability of detecting the surface defects in real time in an industrial production link is improved.

The invention discloses an analysis method and system for a dynamic recrystallization structure form of a titanium alloy based on machine learning and a medium, and belongs to the technical field of metal material microstructure quantitative characterization, a Gleeble thermal compression test is processed on a titanium alloy to be tested, multi-modal data is collected, a DRX probability graph is output based on a DRX segmentation model, and then the characteristics of DRX are extracted. And respectively inputting the DRX features into the first-level classifier, carrying out PCA dimension reduction processing, splicing and fusing the prediction probability and the features after dimension reduction, and inputting the spliced and fused features into a second-level classifier to output a two-dimensional DRX segmentation map. Based on the FIB-SEM tomography sequence image, reconstructing a three-dimensional model of the DRX crystal grain so as to carry out consistency verification on the two-dimensional DRX segmentation image; and evaluating a correlation coefficient of the three-dimensional model and the two-dimensional DRX model, and finally carrying out three-dimensional visualization on the space aggregation of the DRX crystal grains. According to the method, the automation degree and efficiency of dynamic recrystallization proportion and type identification are remarkably improved.

A method of determining ecological risk with spatial aggregation and regional imbalance level applicable for use in a river basin which involves geographic information, ecology, environmental engineering and computer technology includes the steps of: dividing a river basin into a plurality of fishnets; obtaining a basin landscape disturbance index and a basin landscape vulnerability index; obtaining a basin ecological risk index based on the basin landscape disturbance index and the basin landscape vulnerability index; and determining and evaluating the basin ecological risk in spatial aggregation and the regional imbalance with reference to a spatial autocorrelation value and a basin ecological risk imbalance value obtained by using the basin ecological risk index. The ecological risk in spatial aggregation and the regional imbalance is output and evaluate so that the ecological risk in river basin can be assessed comprehensively with high evaluation accuracy.

The invention discloses a traffic scene small target detection method based on an adaptive spatial aggregation pyramid, and belongs to the field of computer vision and small target detection, and the method comprises the steps: designing a multi-scale aggregation attention mechanism to enhance the texture, shape and context information of a small target for the problem of weak semantic information of the small target; aiming at the problem of insufficient multi-scale fusion of small targets, designing an adaptive space aggregation pyramid depth fusion multi-size feature map; a small target loss function NEIoU is constructed to solve the problem that small target detection is sensitive to position deviation, and the convergence speed is increased; and meanwhile, a Soft-NMS strategy is adopted to alleviate the problem that small targets are mistakenly suppressed due to IoU calculation deviation or dense arrangement. According to the traffic scene small target detection method based on the adaptive spatial aggregation pyramid, the precision of small target detection is remarkably improved by improving semantic richness, multi-scale fusion and positioning precision; meanwhile, deployment is easy, the operation speed is high, and the requirement for real-time detection is met.

The invention relates to the technical field of remote sensing data analysis, in particular to a land resource management data analysis system. According to the method, the spectral reflectivity change under the earth surface multi-direction illumination is obtained and the reflection gradient response information of the spectral reflectivity change is extracted, so that the spatial difference of the earth surface reflection characteristics at different angles can be more comprehensively presented; according to the method, surface land type division is more accurate in gray continuity and spatial structure expression, so that potential fission boundaries are identified and reasonable section classification is carried out through quantitative comparison and spatial aggregation analysis of structure texture differences in boundary identification between different regions, and the identification accuracy of the boundary identification between different regions is improved. The stability and variability of the interior and boundary structures of the land areas are effectively described, and on this basis, the area boundary is reconstructed and combined with texture and reflection clustering grouping, so that the land type state and the distribution boundary of each area can be finely judged.

The invention relates to a transformer area distributed photovoltaic power ultra-short-term prediction method and system, and belongs to the technical field of distributed photovoltaic power prediction. On the basis of an agglomerate hierarchical clustering algorithm, transformer area photovoltaic in a region is clustered into a sub-region set with homogenized output characteristics, decoupling representation of spatial-temporal characteristics in the sub-regions is realized in combination with a GraphSAGE graph neural network and a Transform encoder model, and finally, a photovoltaic power prediction result of each transformer area in the sub-regions is synchronously output on the basis of the composite spatial-temporal characteristics. Furthermore, a proper migration modeling strategy is constructed based on shared features among the sub-regions, rapid lightweight modeling of each sub-region prediction model is realized, and finally, a region total power prediction result is obtained through spatial aggregation of sub-region prediction power.

The invention relates to the technical field of information retrieval, in particular to an academic supply and demand dynamic sensing method and system based on space-time big data, and the method comprises the following steps: obtaining the path length of a source and a target college, screening migration direction stable labels, recognizing a trend consistent region, evaluating the local transfer capacity, and fusing a high-frequency path region. And extracting behavior track data, performing sequence comparison on the associated nodes, and generating a prediction chain space offset mark set. According to the method, space label units with stable migration characteristics are effectively distinguished by processing enrollment information and population migration trends and constructing a path length filtering and direction section repeating mechanism, and supply and demand migration origin areas are accurately locked in combination with change trends reflected by continuous direction consistency; spatial aggregation judgment is enhanced through a statistical means of a neighborhood capacity proportion and path intersection frequency, the sensitivity and dynamic response capability of a spatial prediction structure are improved, and active sensing and intervention prompting of a supply and demand change trend under multiple space-time dimensions are realized.

The invention discloses an element quantitative fusion-based urban coastal zone landscape evaluation and prediction method, which comprises the following steps of delimiting a vector boundary of a target urban coastal zone based on land change data, and dividing an area covered by the vector boundary into a plurality of grid units; associating the preprocessed spatial element data to the corresponding grid units, and constructing a structured spatial database; establishing a multi-element quantitative index system, calculating a combined weight, and outputting a comprehensive evaluation score of each grid unit through weighted summation; calculating global and local Moran's I indexes based on the comprehensive evaluation score, and identifying a spatial aggregation region and an abnormal region with statistical significance; obtaining dynamic monitoring data in a coastal zone range of a target urban area, and updating the structured spatial database; and based on the updated structured spatial database, comprehensively evaluating the score and the recognition results of the spatial accumulation area and the abnormal area, and generating an evaluation and prediction result.

The invention discloses a wrist image fracture recognition detection method and system based on an improved YOLOv8 network, and the method comprises the steps: obtaining a to-be-processed wrist fracture image data set, and carrying out the data preprocessing, and obtaining a wrist fracture image; based on a fast network backbone structure, a bidirectional feature pyramid network structure fused with a global-to-local space aggregation module and a dynamic cross-stage deformable convolutional network module is introduced, and an improved YOLOv8 recognition network model is constructed; and performing image recognition processing on the wrist fracture image based on the improved YOLOv8 recognition network model to obtain a wrist image fracture recognition result. According to the invention, the capture capability and feature extraction efficiency of the wrist fracture features under the complex background can be enhanced, and then the precision of wrist fracture image recognition is improved. The wrist image fracture recognition and detection method and system based on the improved YOLOv8 network can be widely applied to the technical field of image recognition.

The invention discloses a defect local resonant frequency extraction method based on vibration response space clustering analysis, and belongs to the field of nondestructive testing. The method comprises the following steps: processing an ultrasonic detection data set, calculating a difference function between a frequency spectrum of each scanning point and an average frequency spectrum, extracting a local maximum frequency, carrying out frequency statistics, and screening out an abnormal response frequency; in combination with the spatial distribution characteristics of the detection points, identifying a frequency cluster with spatial aggregation by adopting a density clustering algorithm to form a candidate local resonant frequency set; and constructing an intensity matrix under the candidate local resonant frequency, extracting energy attenuation characteristics in each direction by using an equal-angle scanning method, and comparing the energy attenuation characteristics with a preset threshold value to determine the defect local resonant frequency. The method does not need to preset defect positions and geometric features, can realize automatic extraction of the defect local resonant frequency under the condition of unknown defects, has the advantages of high identification precision, strong anti-interference capability, high calculation efficiency and the like, and is suitable for non-destructive testing requirements in an engineering environment.

The invention provides an intelligent construction method for a water conservation forest project in a loess plateau area, and relates to the field of forestry ecological restoration and water and soil conservation projects. The method comprises the following steps: collecting and carrying out normalization processing on various original physical quantities based on a space-time sliding window to obtain normalized values of the various physical quantities; constructing a composite physical state function based on the normalized values of various physical quantities, introducing a disturbance propagation response function to construct a measurement evaluation function, and performing spatial threshold screening processing based on the measurement evaluation function to generate a planting candidate point set; generating a spatial aggregation planting block set based on the planting candidate point set; and determining planting parameters based on the gradient change condition of the measurement evaluation function in the spatial aggregation planting block so as to execute planting operation. The problems that a traditional water conservation forest engineering construction method lacks modeling of spatial disturbance and neighborhood consistency, construction parameter configuration lacks intelligent adjustment, and a measurement-analysis-decision-execution closed loop structure is not formed in the whole process are solved.

The invention discloses an aquatic seedling counting and growth parameter measuring system, which comprises the following steps: acquiring a seedling flow image sequence and generating an image confidence score, selecting kernel function width according to different confidence conditions to construct a density map, and obtaining a density distribution feature vector capable of representing a spatial aggregation state; extracting morphological features of the seedlings under a multi-scale receptive field based on the feature vectors, constructing a multi-scale feature pyramid, calculating a spatial overlapping degree index according to a difference relationship between scales, and identifying a dense region; the fusion weight of the multi-scale features is determined in combination with an overlapping degree index, and fusion features are generated and used for executing counting and growth parameter measurement; and constructing a performance deviation index according to the counting accuracy and the measurement precision, and performing joint updating on the kernel function parameters and the fusion weight when the deviation exceeds a threshold value to form an optimized path of which the parameters can be iteratively updated so as to realize stable counting and accurate measurement under complex density change and overlapping conditions.

The invention discloses a traffic anomaly detection method based on a space-time double-flow network and multi-modal feature fusion, relates to the technical field of intelligent video analysis and traffic behavior recognition, and is used for solving the problem of high false detection rate of existing traffic anomaly detection in a complex environment. A video image sequence is extracted through a fixed and dynamic combined frame sampling strategy, and motion features are constructed. And in combination with the direction consistency coefficient and the perturbation evolution trend, repairing direction perception distortion by using a mirror image mapping mechanism. Target appearance features are extracted through spatial branches, behavior evolution trajectories are extracted through time branches, and meanwhile, a modal weight dynamic adjustment mechanism is introduced to enhance the adaptability to complex scenes. The multi-target interaction relation is quantified based on a graph structure modeling mode, an adjacent structure is reconstructed through a graph self-encoder, and local behavior mutation is recognized. And finally realizing high-confidence anomaly marking by combining state transition analysis and spatial aggregation characteristics. The method has the advantages of high real-time performance, high robustness and wide scene adaptability.

The invention discloses an urban spatio-temporal data completion method based on deformable convolution and induction graph aggregation, which comprises the following steps: constructing a sub-graph through sampled nodes, obtaining a missing mask according to a mask matrix, and adding missing information into training data; modeling the time sequence relation by adopting deformable time convolution, and capturing multi-scale time sequence characteristics; and multi-order neighborhood information fusion is carried out through a spatial aggregation network, node characteristics and aggregated neighbor characteristics are connected in series, and data are complemented through multi-loss function joint optimization. A spatial aggregation network is utilized to capture a spatial dependency relationship to realize induction and completion, and model training and reasoning under dynamic change of the number of nodes are realized.

The invention relates to the technical field of three-dimensional image processing, in particular to a distributed parallel processing and loading method for large-scale scene three-dimensional model data, which comprises the following steps of: acquiring a plurality of large-scale scene three-dimensional model patch coordinate sets, normal vector sets and material identification sets, calculating an adjacent matrix and judging a spatial aggregation region; the method comprises the following steps: segmenting and generating regional data subsets, calculating patch quantity and texture size matching calculation node capability, constructing a distribution path table, extracting a continuous triangular patch chain and reconstructing compressed data, calculating a loading priority in combination with a view angle, and loading and caching data according to scheduling. The spatial aggregation degree of adjacent patches is calculated, spatial regions are divided by combining material change points, the computing power and bandwidth capacity of nodes are dynamically matched and computed according to the patch number and the texture size, and the screen space intersection area and the viewpoint distance are computed by combining the user view angle to adjust the loading priority. And the visual angle-oriented visual scheduling response efficiency is improved.

The application provides a GH4169 forging grain structure ultrasonic evaluation method positioned along sound path zoning, extracts frequency spectrum centroid and high frequency / low frequency energy ratio as frequency domain characteristic parameters through wavelet packet decomposition, fully utilizes the physical law of high frequency scattering enhancement of coarse grain area, and has higher organization state distinguishing sensitivity compared with traditional time domain statistics; the application also provides a double threshold value scoring positioning method based on statistical baseline and an unsupervised positioning method based on K-means clustering, the overlapping identification rate of the two methods in coarse grain dominant area reaches more than 90%, and the two methods can be flexibly selected or verified according to the presence or absence of prior knowledge, and the reliability of the detection result is improved; the sliding window density criterion in the application realizes the judgment from single point anomaly to area scale by investigating the spatial aggregation of abnormal windows along the depth direction, effectively avoids false alarm caused by isolated noise points, and significantly improves the robustness of the detection result.

The invention relates to the technical field of medical image processing, and discloses a chronic obstructive pulmonary emphysema distribution quantitative method and system based on multi-scale segmentation, and the method comprises the steps: carrying out the anisotropic diffusion filtering noise reduction of a chest CT image; segmenting a lung field and removing a blood vessel bronchial structure by adopting a region growing algorithm; multi-scale image representation is constructed based on a Gaussian pyramid, an emphysema candidate area is identified in a coarse scale layer, and a boundary is accurately drawn by adopting a self-adaptive threshold value in a fine scale layer; extracting local texture features to distinguish the lobular central emphysema and the total lobular emphysema; dividing severity levels according to spatial aggregation characteristics and density gradient distribution, and calculating an air swelling volume ratio and a distribution heterogeneity index; the three-dimensional pseudo-color volume is used for drawing visualization, a structured quantitative report is generated, accurate segmentation and subtype classification of the emphysema area are achieved, and comprehensive quantitative analysis indexes are provided.

The application provides a water conservancy entity space feature index and efficient rendering method based on machine learning and WebGL, and comprises the following steps: decomposing each water conservancy entity in a watershed into a space feature, and establishing a watershed water conservancy entity space feature index database; establishing a feature vector of each water conservancy entity according to the space feature of each water conservancy entity in the watershed; clustering each water conservancy entity based on the feature vector of each water conservancy entity, obtaining a clustering label of each water conservancy entity, and establishing a watershed water conservancy entity clustering label index database; and rendering the water conservancy entity in the watershed according to the watershed water conservancy entity space feature index database and the watershed water conservancy entity clustering label index database. The application has the following advantages: the established watershed water conservancy entity space feature index database and the watershed water conservancy entity clustering label index database can efficiently realize visual expression, spatial aggregation and classified retrieval in a GIS platform, a database or an application system.

The present application relates to the technical field of artificial intelligence, and more particularly to an AI interaction system for ornithological scientific research, which comprises a collection unit, a trigger determination unit, a path determination unit, a dominant determination unit, a prompt generation unit and an adjustment unit. The present application constructs a multi-level behavior determination process through joint analysis of bird song spectrum, image motion characteristics, flight trajectory and environmental factors, and through statistical analysis of the time distribution and spatial distribution of these behavior indicators within a preset period, and correlation analysis with environmental factors such as temperature, humidity, wind speed and illumination, the influence of environmental changes on bird activity frequency and spatial aggregation degree can be identified, so that the acoustic trigger threshold can be dynamically adjusted, so that the system can continuously maintain the sensitive recognition ability of bird activity with the change of the habitat environment, and then realize high-reliable behavior recognition and interaction prompt for ornithological scientific research.

The embodiment of the invention discloses a water and soil loss pattern spot comprehensive drawing method based on spatial aggregation degree analysis, and relates to the field of geographic information technology and image data processing. The method comprises the steps that water and soil loss pattern spot vector data of a target area are acquired, rasterization processing is carried out on the water and soil loss pattern spot vector data to obtain raster data corresponding to each raster unit, and the raster data corresponding to each raster unit at least comprises soil erosion information, land utilization type information and DEM information; calculating an aggregation degree characterization parameter corresponding to each grid unit based on the grid data; determining spatial aggregation characteristics of the water and soil loss pattern spots of the target area according to the aggregation degree characterization parameters, and drawing an aggregation degree analysis image of the water and soil loss pattern spots according to the spatial aggregation characteristics; and performing hierarchical expression and layer creation on the water and soil loss pattern spots by using the aggregation degree analysis image of the water and soil loss pattern spots as a reference basis, and synthesizing a final water and soil loss pattern spot comprehensive graph.

An exemplary multistart and recombination system and method are disclosed for determining unique solutions to spatial aggregation problems using a multistart phase and a recombination phase algorithm, where the multistart phase generates a diverse pool of random solutions and where the recombination algorithm uses the generated random solution to create new, improved solutions, ensuring the discovery of multiple unique optimal or near-optimal configurations. The spatial aggregation problems may be used to analyze census data. In some embodiments, spatial aggregation problems may be used to perform voter district map analysis to identify a revised voter district map.

The application discloses a next interest point recommendation method based on a multi-active area perception network. The method fully utilizes the spatial aggregation of user behavior trajectories, filters the active area center of the user by analyzing the historical activity trajectory of the user, and divides and aggregates according to the distance to realize the acquisition of long-term stable user preference features in the active area. At the same time, the method is based on a negative sample sampling method of the neighbor area, filters the interest point through the activity area of the user, fully utilizes the geographical location characteristics, and acquires sample information with more rich information. The application is based on historical check-in data and other multi-modal information of the user, constructs a network model through a deep learning method, extracts user preference features to predict the next most possible place visited by the current user, has the characteristics of high accuracy and strong scalability, can timely grasp the user behavior trend, and provides guidance for user behavior trajectory prediction personnel.

The invention discloses an airborne ocean laser radar hierarchy and water bottom synchronous three-dimensional reconstruction method. The method comprises the steps of obtaining a three-dimensional point cloud of laser on a water surface according to airborne laser radar data; gridding is carried out on the three-dimensional point cloud of the laser on the water surface, multi-pulse averaging of spatial aggregation is carried out, and signals after gridding aggregation are obtained; and obtaining a water surface peak position, a level peak position and a water bottom peak position, and finally obtaining water bottom and level three-dimensional point clouds at the same time, thereby realizing high-precision three-dimensional reconstruction. According to the method, the thought of voxelized grids is combined, resolution sacrifice caused by multi-pulse averaging in a large space range is avoided, high-resolution three-dimensional reconstruction with adjustable resolution is achieved, a multi-target distinguishing algorithm architecture is designed, and the effect of multi-target simultaneous detection is achieved; the method has reference significance for shallow sea environment detection and shallow sea carbon cycle analysis, and has wide applicability.

The present application relates to a kind of table area distributed photovoltaic power ultra-short term prediction method and system, belong to distributed photovoltaic power prediction technical field.Based on the condensed hierarchical clustering algorithm, the table area photovoltaic in region is clustered into the homogeneous sub-region set of output characteristics, combined with GraphSAGE graph neural network and Transformer encoder model, the decoupling representation of space-time characteristics in sub-region is realized, based on composite space-time characteristics, finally synchronously output the photovoltaic power prediction result of each table area in sub-region, further based on the shared feature between sub-region, construct suitable migration modeling strategy, realize the fast lightweight modeling of each sub-region prediction model, finally, the spatial aggregation of sub-region prediction power obtains regional total power prediction result.

The present invention relates to the technical field of electrical digital data processing, and in particular to a method and system for remote sensing hyperspectral sharpening and fidelity guided by spectral and spatial aggregation. The method includes the following steps: obtaining hyperspatial pixels, a spatial pixel affinity matrix, hyperspectral channels, and a spectral affinity matrix; performing information interaction to obtain the interaction features of the hyperspatial pixels and the interaction features of the hyperspectral channels; performing self-attention interaction to obtain the updated hyperspectral features of each hyperspectral channel and the updated hyperspectral features; and fusing the updated hyperspatial features with the updated hyperspectral features to obtain a remote sensing hyperspectral sharpened and fidelity image. The method and system provided by the present invention achieve an improvement in spatial resolution while ensuring the integrity of the spectral dimension through spatial and spectral feature aggregation and interaction mechanisms, providing strong technical support for the analysis and practical application of remote sensing hyperspectral images.

The invention provides an automatic driving target detection method based on improved YOLOv8. The automatic driving target detection method comprises the steps of collecting an automatic driving scene image, labeling a target, generating a bounding box and a category label, randomly dividing a data set into a training set, a verification set and a test set according to a proportion of 8: 1: 1, and performing data enhancement on the data set; a basic YOLOv8n model is improved, and a C2f module in a trunk part is replaced by a C2fDCNv4 module; the Neck part uses a self-attention mechanism module COT to replace a third C2f module and a fourth C2f module; in the Head part, an LSDCH lightweight shared convolution detection head is used for replacing a detection head of the YOLOv8n; and training and evaluating the improved YOLOv8n model, and obtaining the YOLOv8n model for target detection after the training and evaluation are passed. According to the method, softmax normalization in space aggregation is removed, the modulation scalar is converted into unbounded dynamic weight, so that the convolution kernel can adaptively adjust the receptive field according to the shape and attitude of the target, the calculation efficiency and performance are balanced through lightweight design, and the method is suitable for edge deployment.

The application provides a kind of urban dissipative space discrimination method and system based on spatial aggregation characteristics, it is related to spatial quality measure technical field.The urban dissipative space discrimination method based on spatial aggregation characteristics, comprising: obtaining city basic geographic space data and its geographic coordinates, form city space information integrated platform;According to the pre-set city dissipative space determination index library, the basic characteristic element data of city dissipative space is calculated, and the basic characteristic element data of city dissipative space is loaded into city space information integrated platform for spatial positioning calibration;Select appropriate grid scale to divide measure area;Each characteristic element data in grid unit is graded assignment;The order of city space is measured to calculate the aggregation;Establish determination rule to obtain four kinds of dissipative space characteristic element data set.The application improves the accuracy and efficiency of city dissipative space identification, and provides targeted guidance and power point for city renewal design and decision control.

This invention discloses a method for creating fish habitats based on swarming effects, belonging to the fields of fish ecological engineering and water conservancy engineering. Existing technologies for creating fish habitats suffer from incomplete quantification of swarming behavior and a lack of quantitative correlation with engineering design parameters, leading to habitats that are difficult to match the needs of fish swarming. This invention quantifies spatial aggregation, movement synchronization, and turbulent adaptability, calculates a comprehensive swarming index and a group spatial demand coefficient, establishes a direct conversion model between swarming behavior indicators and habitat design parameters, and determines flow field control methods and structural layout methods based on the comprehensive swarming index and the group spatial demand coefficient, thereby creating fish habitats. This invention can comprehensively characterize swarming behavior, scientifically guide engineering design, and improve habitat applicability and utilization.

This application discloses a method for parallel integration of geospatial data based on spatial grid-aware routing. The method includes: acquiring source information of geospatial data files and identifying their format and metadata; logically segmenting the data files into record blocks; calculating the bounding rectangle of the geometric objects of geospatial elements in each record block and determining the covering grid set according to spatial grid partitioning rules, then determining a unique master spatial grid identifier according to master attribution rules; constructing grid clusters based on the master spatial grid identifier, merging adjacent grids into scheduling partitions and routing them to corresponding parallel processing tasks; and having each task stream-read and process elements before batch writing them to the target storage. This invention overcomes the bottleneck of single-file parallel processing, achieves spatial locality-aware routing and dynamic load balancing, improves the spatial aggregation and query performance of target data, reduces the risk of memory overflow, and possesses high-throughput writing and precise breakpoint resumption capabilities.

This invention relates to the field of electronic digital data processing technology and discloses a method for processing tunnel construction data based on real-time monitoring. The method involves acquiring real-time data from displacement, stress, and micro-vibration sensors; constructing a spatial graph adjacency matrix based on coordinates and rock strata topology; and plotting the dynamic features of nodes in the time series graph. The data is then input into a physically constrained spatiotemporal graph convolutional network. A displacement-stress transmission constraint term based on the elasticity equation is introduced as a weight attenuation penalty factor in the graph convolutional propagation layer to restrict cross-node information transmission in accordance with the laws of surrounding rock mechanics. The dynamic features of the time dimension are extracted through a time-gated cyclic unit, and a tensor data stream is output. This invention restricts weight updates within physical boundaries, eliminates non-physical interference features from spatial aggregation, avoids feature confusion during multi-source data fusion, and overcomes the problem of distorted precursor feature extraction due to a lack of physical constraints. Furthermore, improvements are made to the coupling processing of multi-source heterogeneous data under geological topology, taking into account classification numbers.