Augmented reality (ar) assisted ecological pollution source localization method and system

Abstract

The present application relates to the technical field of ecological pollution source positioning, and particularly relates to an AR-assisted ecological pollution source positioning method and system. Based on the pollutant multi-medium interphase migration characteristics and the concentration gradient space-time correlation mode, a quality exchange network is constructed, and the pollution source is traced back to obtain a source feature set containing spatial coordinates and medium contribution weight; according to the spatial coordinates, the ground feature of the AR field of view is extracted and matched with the satellite remote sensing image, and the source feature set is converted into a three-dimensional virtual labeling object; the user-selected field verification point and the measured concentration value are received, a residual distribution field is constructed, and a spatial abnormal area is identified, and accordingly the medium contribution weight and the migration characteristics are adjusted to re-calculate the positioning result. The method improves the accuracy and interactivity of the pollution source positioning.

Description

Technical Field

[0001] This invention relates to the field of ecological pollution source location technology, and in particular to an AR-assisted ecological pollution source location method and system. Background Technology

[0002] In the field of locating ecological pollution sources, current conventional practices typically rely on traditional environmental monitoring and numerical simulation methods. Monitoring personnel need to establish fixed sampling stations in the target area and regularly collect soil, water, air, and biological samples for laboratory analysis to obtain pollutant concentration data in each medium. This data is then input into numerical simulation systems based on diffusion equations or box models to estimate pollutant transport paths and potential sources. Simultaneously, satellite remote sensing imagery and Geographic Information Systems (GIS) are also used to assist in identifying the correlation between surface pollutant distribution and land use types. However, these methods heavily rely on discrete spatiotemporal sampling data, and the inter-media migration characteristics of pollutants are often simplified to fixed parameters or empirical coefficients, lacking a refined description of dynamic exchange networks.

[0003] Traditional monitoring networks, limited by the number and location of sampling points, struggle to capture instantaneous flux changes of pollutants at the soil-water-atmosphere-biosphere interface. This results in significant spatial ambiguity in reverse source tracing results, making it impossible to accurately pinpoint source coordinates. Numerical simulation results from traditional monitoring networks are typically presented as two-dimensional maps or data tables, lacking a direct mapping to the actual on-site environment. Environmental managers, during field investigations, must manually match abstract pollutant migration paths with complex actual terrain, which is not only inefficient but also prone to overlooking critical transmission nodes due to perspective bias and obstruction from ground features. Furthermore, the lack of systematic spatial residual analysis methods for discrepancies between measured data and simulation predictions hinders dynamic optimization of model parameters, preventing closed-loop correction for pollution source location. These limitations restrict the accuracy and real-time response capability of ecological pollution source location. Summary of the Invention

[0004] The present invention provides an AR-assisted method and system for locating ecological pollution sources, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides an AR-assisted method for locating ecological pollution sources, comprising: Based on the interphase migration characteristics of pollutants in soil-water-atmosphere-biological multi-media and the spatiotemporal correlation patterns of pollutant concentration gradients between different media, a mass exchange network for pollutants across phase interfaces is constructed. This mass exchange network identifies the spatiotemporal correlation patterns of pollutant concentration gradients in different media in heterogeneous environmental data of the target ecological region, and traces the transport path of pollutants from the receptor point to the source in reverse, obtaining a pollution source tracing feature set containing spatial coordinate information and media contribution weights. Based on the spatial coordinate information, the edge features and texture features of ground objects in the real-time field of view of the AR device are extracted and registered with the corresponding ground object features in the satellite remote sensing image at multiple scales. By constructing a hierarchical mapping relationship from the world coordinate system to the device field of view coordinate system, the pollution source tracing feature set is converted into a three-dimensional virtual annotation object in the AR scene. The system receives the on-site verification points and corresponding measured pollutant concentration values ​​selected by the user through the interactive interface of the three-dimensional virtual annotation object. It constructs a residual distribution field between the measured pollutant concentration values ​​and the predicted concentration values ​​of the pollution source tracing feature set, identifies spatially anomalous regions in the residual distribution field, adjusts the medium contribution weight and the interphase migration characteristics according to the distribution characteristics of the spatially anomalous regions, recalculates the pollution source tracing feature set, and obtains the ecological pollution source location result.

[0006] Based on the interphase migration characteristics of pollutants in multiple media (soil, water, atmosphere, and organisms) and the spatiotemporal correlation patterns of pollutant concentration gradients between different media, a mass exchange network for pollutants across phase interfaces is constructed, including: By extracting soil microbial community structure data from the heterogeneous environmental data, the spatial distribution gradient of specific degrading bacteria was analyzed, and the spatial coupling relationship between changes in degrading bacteria density and changes in pollutant concentration was identified. Extract vegetation spectral response data from the heterogeneous environmental data, identify the spatial aggregation pattern and spectral reflectance shift amplitude of chlorophyll fluorescence anomaly regions, and determine the accumulation path and residence time characteristics of pollutants in biological media based on the spatial coupling relationship, the spatial aggregation pattern and the spectral reflectance shift amplitude. The accumulation path and residence time characteristics are cross-validated with the spatiotemporal distribution data of surface pollutant concentration and groundwater flow vector field data in the heterogeneous environmental data. By comparing the time series of pollutant accumulation in biological media with the time series of pollutant migration in non-biological media, the spatiotemporal correlation patterns of pollutant concentration gradients between different media are identified. Based on the spatiotemporal correlation pattern and the interphase migration characteristics of pollutants in soil-water-atmosphere-biological multi-media, the mass exchange network spanning abiotic and biological media is constructed.

[0007] The quality exchange network identifies the spatiotemporal correlation patterns of pollutant concentration gradients in different media within heterogeneous environmental data of the target ecological region, and traces the transport path of pollutants from receptor points to sources in reverse, obtaining a pollution source tracing feature set containing spatial coordinate information and media contribution weights, including: The ratio of the cumulative value of pollutant concentration gradient along the path direction to the path length is calculated as a concentration gradient intensity index, and the ratio of the time difference between the peak pollutant concentration along the path direction and the starting point of the path to the path length is calculated as a time propagation rate index. Based on the concentration gradient intensity index and the time propagation rate index, the spatiotemporal consistency score of each path is determined, and the dominant transmission path with a spatiotemporal consistency score exceeding the preset path screening threshold is selected. The pollutant is traced backward from the receptor point to the source along the dominant transport path. The starting position of the dominant transport path is identified as the spatial coordinate information of the pollution source. The mass transport flux information of each medium node passed through by the dominant transport path is extracted. The contribution of each medium to the pollution source is calculated according to the proportion of the cumulative mass of pollutants transported from each medium node to the pollution source location to the total transport mass. The medium contribution weight is generated. The spatial coordinate information and the medium contribution weight are combined to form the pollution source tracing feature set.

[0008] Based on the spatial coordinate information, the edge and texture features of ground objects in the real-time field of view of the AR device are extracted, and multi-scale registration is performed with the corresponding ground object features in the satellite remote sensing image, including: Based on the spatial coordinate information, the geographical location of the pollution source in the world coordinate system is determined, the field of view of the AR device is obtained, and the edge features and texture features of ground objects within the field of view are extracted. Remote sensing images covering the geographical location are retrieved from the satellite remote sensing image database. Corresponding ground features in the remote sensing images corresponding to the field of view are extracted. By constructing a multi-scale feature pyramid, the edge and texture features of ground features within the field of view are registered layer by layer with the corresponding ground features in the remote sensing images at different spatial resolution levels. At the coarsest resolution level of the multi-scale feature pyramid, the spatial position offset and rotation angle offset between the edge features of ground objects within the field of view and the corresponding edge features of ground objects in the remote sensing image are calculated. The spatial position offset and the rotation angle offset are used as initial registration parameters and passed to the next resolution level. At the next resolution level, the local deformation parameters between texture features are further calculated based on the initial registration parameters. The registration parameters are passed layer by layer until the finest resolution level, resulting in multi-level registration parameters from the world coordinate system to the device field of view coordinate system.

[0009] By constructing a hierarchical mapping relationship from the world coordinate system to the device field-of-view coordinate system, the pollution source tracing feature set is converted into a three-dimensional virtual annotation object in the AR scene, including: Based on the multi-level registration parameters, a hierarchical mapping relationship from the world coordinate system to the device field of view coordinate system is constructed. The hierarchical mapping relationship includes a global coordinate transformation matrix and a local deformation field. The global coordinate transformation matrix is ​​used to transform the spatial coordinate information from the world coordinate system to the camera coordinate system of the AR device. The local deformation field is used to compensate for local geometric distortions caused by differences in viewing angle and terrain undulations between the field of view image and the remote sensing image. The spatial coordinate information and the medium contribution weight in the pollution source tracing feature set are transformed to the device field of view coordinate system through the hierarchical mapping relationship. Based on the transformed coordinate information, a three-dimensional geometric model identifying the location of the pollution source is generated in the AR scene. Based on the medium contribution weight, visual encoding attributes are assigned to different parts of the 3D geometric model. The visual encoding attributes include color mapping and transparency mapping. The color mapping determines the display color of the corresponding part of the 3D geometric model according to the magnitude of the medium contribution weight. The transparency mapping determines the transparency value of the corresponding part of the 3D geometric model according to the confidence level of the medium contribution weight, thereby generating the 3D virtual annotation object.

[0010] Receiving the on-site verification points and corresponding measured pollutant concentration values ​​selected by the user through the interactive interface of the 3D virtual annotation object, and constructing the residual distribution field between the measured pollutant concentration values ​​and the predicted concentration values ​​of the pollution source tracing feature set includes: The system receives touch operation commands from users on the interactive interface of the three-dimensional virtual annotation object, and marks the three-dimensional spatial position of the on-site verification point in the AR scene according to the touch operation commands. The three-dimensional spatial position is converted into geographic coordinates in the world coordinate system through the inverse mapping of the hierarchical mapping relationship. Receive the measured pollutant concentration values ​​collected by the user at the on-site verification point; The predicted concentration values ​​at the corresponding geographical coordinates of the field verification points are extracted from the pollution source tracing feature set. The predicted concentration values ​​are calculated by the mass exchange network based on the interphase migration characteristics and the medium contribution weight. For each medium at the field verification point, calculate the concentration residual value between the measured pollutant concentration value and the predicted concentration value. When there are multiple field verification points, summarize the concentration residual values ​​of all field verification points to form a residual dataset. Using the geographic coordinates of each field verification point in the residual dataset as spatial anchors, the discrete concentration residual values ​​are extended into a continuously distributed residual distribution field within the target ecological area through spatial interpolation. The spatial interpolation method calculates the residual estimate of the unsampled location based on the distance decay principle. The residual estimate gradually decays to zero as the distance between it and the nearest field verification point increases.

[0011] Identify spatially anomalous regions in the residual distribution field, and adjust the medium contribution weight and interphase migration characteristics based on the distribution characteristics of the spatially anomalous regions. Recalculate the pollution source tracing feature set to obtain the ecological pollution source location results, including: Calculate the global residual statistical characteristics of the residual distribution field; determine the residual anomaly judgment threshold based on the global residual statistical characteristics; mark grid nodes in the residual distribution field whose absolute value of the residual estimate exceeds the residual anomaly judgment threshold as abnormal nodes; and identify the connected regions formed by spatially adjacent abnormal nodes as spatial abnormal regions. The geometric center location and residual amplitude characteristics of the spatial anomaly region are extracted as distribution features; The geometric center of the spatial anomaly region is mapped to the node space of the quality exchange network, the medium type corresponding to the geometric center is identified, the adjustment direction of the medium contribution weight is determined according to the positive or negative sign of the residual amplitude feature, and the adjustment magnitude of the medium contribution weight is determined according to the average amplitude of the residual amplitude feature. The phase migration characteristics between the medium type and adjacent medium types are adjusted according to the coverage area and boundary shape of the spatial anomaly region. The adjusted medium contribution weight and the adjusted interphase migration characteristics are input into the mass exchange network, and the reverse tracing calculation of the transmission path from the receptor point to the source is re-executed to obtain the updated pollution source tracing feature set. The spatial coordinate information in the updated pollution source tracing feature set is used as the ecological pollution source location result.

[0012] A second aspect of the present invention provides an AR-assisted ecological pollution source localization system, comprising: The pollution source tracing unit is used to construct a mass exchange network of pollutants across phase interfaces based on the interphase migration characteristics of pollutants in soil-water-atmosphere-biological multi-media and the spatiotemporal correlation patterns of pollutant concentration gradients between different media. The mass exchange network traces the transmission path of pollutants from the receptor point to the source by identifying the spatiotemporal correlation patterns of pollutant concentration gradients in different media in the heterogeneous environmental data of the target ecological area, and obtains a pollution source tracing feature set containing spatial coordinate information and media contribution weights. The processing unit is used to extract the edge features and texture features of ground objects in the real-time field of view of the AR device based on the spatial coordinate information, perform multi-scale registration with the corresponding ground object features in the satellite remote sensing image, and convert the pollution source tracing feature set into a three-dimensional virtual annotation object in the AR scene by constructing a hierarchical mapping relationship from the world coordinate system to the device field of view coordinate system. The processing unit is used to receive the on-site verification points and corresponding measured pollutant concentration values ​​selected by the user through the interactive interface of the three-dimensional virtual annotation object, construct the residual distribution field between the measured pollutant concentration values ​​and the predicted concentration values ​​of the pollution source tracing feature set, identify spatially anomalous regions in the residual distribution field, adjust the medium contribution weight and the interphase migration characteristics according to the distribution characteristics of the spatially anomalous regions, recalculate the pollution source tracing feature set, and obtain the ecological pollution source location result.

[0013] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] The beneficial effects of this application are as follows: This method, by constructing a mass exchange network across multiple media, can trace the complete transport path of pollutants from the receiver to the source, significantly improving the accuracy of pollution source location. Traditional monitoring methods are often limited to a single medium, making it difficult to capture the inter-media migration behavior of pollutants among soil, water, atmosphere, and organisms, resulting in blind spots in the source tracing results. This scheme utilizes the spatiotemporal correlation pattern of pollutant concentration gradients to automatically identify mass exchange nodes between different media, enabling the location results to comprehensively reflect the real pollution diffusion path of complex ecosystems, avoiding the biases caused by single-medium monitoring, and is particularly suitable for complex pollution scenarios with multiple interfaces.

[0016] Converting pollution source tracing feature sets into 3D virtual labeled objects in AR scenes greatly enhances the intuitiveness and operational efficiency of on-site investigations. Through multi-scale registration and hierarchical mapping, users can directly see the virtual labeled locations of pollution sources and the contribution weights of media in the real environment without professional knowledge, lowering the threshold for data interpretation. At the same time, the AR interactive interface allows users to easily select verification points and input measured data, comparing the differences between predicted and measured values ​​in real time. This transforms the on-site verification process from passive observation to active interaction, shortening the verification cycle caused by the disconnect between traditional laboratory analysis and on-site work.

[0017] A feedback adjustment mechanism is constructed based on the residual distribution field between measured and predicted values, enabling dynamic optimization of pollution source location results. Traditional methods struggle to correct model parameters once determined, while this approach identifies spatial anomalies in the residual distribution field and automatically adjusts the medium contribution weights and interphase migration characteristics, ensuring the source tracing model continuously adapts to the actual field conditions. This closed-loop iterative mechanism effectively eliminates location errors caused by inaccurate initial model parameters or environmental heterogeneity. With repeated user verification and parameter corrections, the location results gradually converge to the true pollution source location, significantly improving the reliability of location in complex terrain and sudden pollution scenarios. Attached Figure Description

[0018] Figure 1 A flowchart illustrating the AR-assisted method for locating ecological pollution sources; Figure 2 A schematic diagram illustrating the process of converting pollution source traceability feature sets into AR 3D virtual annotation objects. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 This is a flowchart illustrating the AR-assisted method for locating ecological pollution sources according to an embodiment of the present invention.

[0022] AR-assisted methods for locating ecological pollution sources include: Based on the interphase migration characteristics of pollutants in soil-water-atmosphere-biological multi-media and the spatiotemporal correlation patterns of pollutant concentration gradients between different media, a mass exchange network for pollutants across phase interfaces is constructed. This mass exchange network identifies the spatiotemporal correlation patterns of pollutant concentration gradients in different media in heterogeneous environmental data of the target ecological region, and traces the transport path of pollutants from the receptor point to the source in reverse, obtaining a pollution source tracing feature set containing spatial coordinate information and media contribution weights. Based on the spatial coordinate information, the edge features and texture features of ground objects in the real-time field of view of the AR device are extracted and registered with the corresponding ground object features in the satellite remote sensing image at multiple scales. By constructing a hierarchical mapping relationship from the world coordinate system to the device field of view coordinate system, the pollution source tracing feature set is converted into a three-dimensional virtual annotation object in the AR scene. The system receives the on-site verification points and corresponding measured pollutant concentration values ​​selected by the user through the interactive interface of the three-dimensional virtual annotation object. It constructs a residual distribution field between the measured pollutant concentration values ​​and the predicted concentration values ​​of the pollution source tracing feature set, identifies spatially anomalous regions in the residual distribution field, adjusts the medium contribution weight and the interphase migration characteristics according to the distribution characteristics of the spatially anomalous regions, recalculates the pollution source tracing feature set, and obtains the ecological pollution source location result.

[0023] In one optional implementation, based on the interphase migration characteristics of pollutants in multiple media (soil-water-atmosphere-biological) and the spatiotemporal correlation patterns of pollutant concentration gradients between different media, a mass exchange network for pollutants across phase interfaces is constructed, including: By extracting soil microbial community structure data from the heterogeneous environmental data, the spatial distribution gradient of specific degrading bacteria was analyzed, and the spatial coupling relationship between changes in degrading bacteria density and changes in pollutant concentration was identified. Extract vegetation spectral response data from the heterogeneous environmental data, identify the spatial aggregation pattern and spectral reflectance shift amplitude of chlorophyll fluorescence anomaly regions, and determine the accumulation path and residence time characteristics of pollutants in biological media based on the spatial coupling relationship, the spatial aggregation pattern and the spectral reflectance shift amplitude. The accumulation path and residence time characteristics are cross-validated with the spatiotemporal distribution data of surface pollutant concentration and groundwater flow vector field data in the heterogeneous environmental data. By comparing the time series of pollutant accumulation in biological media with the time series of pollutant migration in non-biological media, the spatiotemporal correlation patterns of pollutant concentration gradients between different media are identified. Based on the spatiotemporal correlation pattern and the interphase migration characteristics of pollutants in soil-water-atmosphere-biological multi-media, the mass exchange network spanning abiotic and biological media is constructed.

[0024] To acquire heterogeneous environmental data for the target ecological region, specific extraction and analysis were conducted on soil microbial community structure data in biological media. Soil microbial community structure data originated from high-throughput sequencing results from field sampling points and real-time monitoring records from in-situ sensor arrays, covering the species abundance, relative proportion, and spatial coordinates of degrading bacteria at different soil depths. By spatially interpolating the density values ​​of degrading bacteria at each sampling point, a continuous spatial distribution field of bacterial density was constructed, and the rate of change of degrading bacterial density within adjacent sampling intervals was calculated to form a bacterial density gradient matrix. Simultaneously, the pollutant concentration gradient distribution at corresponding spatial locations was extracted from soil pollutant concentration monitoring data for the same region. Through spatial overlay analysis, the spatial coupling relationship between changes in degrading bacterial density and changes in pollutant concentration was identified. Specifically, in areas with large pollutant concentration gradients, if the density of degrading bacteria simultaneously shows high-value aggregation, it indicates the existence of microbial-mediated pollutant degradation hotspots in that area. This spatial coupling relationship was recorded as the basis for the bioactivity weighting in subsequent multi-media migration modeling.

[0025] After identifying the spatial coupling relationships of microorganisms, vegetation spectral response data was further extracted from heterogeneous environmental data. This data was derived from multispectral or hyperspectral remote sensing images, focusing on the radiant intensity and red-edge reflectance characteristics of the chlorophyll fluorescence band (typically concentrated around 685nm and 740nm). Pixel-by-pixel anomaly detection of the vegetation canopy spectral reflectance identified sets of pixels with chlorophyll fluorescence intensity significantly lower than the regional background mean. Spatial clustering analysis was performed on these anomalous pixels to extract spatial aggregation patterns of chlorophyll fluorescence anomaly regions, including structural parameters such as the area, shape index, and spatial connectivity of anomalous patches. Simultaneously, the spectral reflectance shift ΔR of each anomalous pixel relative to a healthy vegetation reference area was calculated. ΔR is defined as the difference between the reflectance of the target pixel in a specific band and the mean of the reference area; a larger shift indicates a deeper degree of pollutant stress on the vegetation. Based on the aforementioned spatial coupling relationships among microorganisms, high-value regions of the degradation microbial community density gradient are spatially overlaid and compared with regions of abnormal chlorophyll fluorescence to determine the accumulation pathways of pollutants in biological media, i.e., the preferential transport channels for pollutants from soil through root absorption into vegetation. Furthermore, by analyzing the temporal series changes of spectral reflectance shift amplitude ΔR in remote sensing images from different time phases, and combining this with vegetation growth cycle characteristics, the residence time characteristics of pollutants in biological media are estimated, including the average residence time of pollutants within vegetation and the peak accumulation time points.

[0026] The cumulative pathways and residence time characteristics obtained from biological media analysis were cross-validated with spatiotemporal distribution data of surface pollutant concentrations and groundwater flow vector field data from heterogeneous environmental data. The spatiotemporal distribution data of surface pollutant concentrations were derived from historical records of ground monitoring stations and mobile monitoring platforms, stored as time-series raster data. The groundwater flow vector field data were obtained from hydrogeological borehole data and groundwater level dynamic monitoring data through hydrological simulation calculations, describing the velocity and direction distribution of groundwater in three-dimensional space. During cross-validation, the phase relationship between the time series of pollutant accumulation in biological media and the time series of pollutant migration in abiotic media (soil, water bodies) was compared. If the peak pollutant concentration in biological media lagged behind the peak pollutant concentration in soil or groundwater in time, and the lag duration matched the plant root absorption kinetic parameters, the rationality of the bioaccumulation pathway was verified. If there was a significant deviation between the two in time series, it suggested the existence of unidentified pollutant migration channels or interference from biodegradation processes. By statistically analyzing the time series comparison results of multiple verification sites, the spatiotemporal correlation patterns of pollutant concentration gradients between different media were extracted, including typical time delay parameters of concentration gradient transmission between various media. (in It indicates the characteristic time delay, spatial attenuation law, and dominant transport direction of pollutants migrating from one medium to an adjacent medium.

[0027] Based on the identified spatiotemporal correlation patterns, and combined with the interphase migration characteristics of pollutants in soil-water-atmosphere-biological multi-media, a mass exchange network spanning abiotic and biological media is constructed. This network uses pollutant concentration nodes in each medium as vertices and mass fluxes at different medium interfaces as directed edges. The exchange rate of pollutants between adjacent media is described by quantifying the mass transfer coefficients at each interface. At the abiotic level, mass exchange at the soil-water interface is characterized by percolation flux and adsorption-desorption equilibrium parameters; mass exchange at the water-atmosphere interface is characterized by the volatiles rate constant and Henry's law constant; and mass exchange at the atmosphere-soil interface is characterized by dry and wet deposition fluxes. At the biological level, mass exchange at the soil-vegetation interface is characterized by root absorption rate and bioaccumulation factors. The previously extracted cumulative path and residence time characteristics are used as constraints to correct the mass input and output fluxes of biological media nodes. The bioactivity weights obtained from the microbial spatial coupling analysis are assigned to the corresponding spatial degradation terms to correct the net flux calculation of pollutants in the soil node. The entire mass exchange network is characterized by characteristic time delay parameters in the time dimension. The step size is dynamically updated to ensure that the network structure reflects the migration state of pollutants in the multi-media system over time. The construction of this network provides a complete multi-media dynamics framework for subsequent reverse tracing of pollutant transport paths and generation of pollution source tracing feature sets.

[0028] In one optional implementation, the quality exchange network identifies spatiotemporal correlation patterns of pollutant concentration gradients in different media within heterogeneous environmental data of the target ecological region, and traces the transport path of pollutants from the receptor point to the source in reverse, obtaining a pollution source tracing feature set containing spatial coordinate information and media contribution weights, including: The ratio of the cumulative value of pollutant concentration gradient along the path direction to the path length is calculated as a concentration gradient intensity index, and the ratio of the time difference between the peak pollutant concentration along the path direction and the starting point of the path to the path length is calculated as a time propagation rate index. Based on the concentration gradient intensity index and the time propagation rate index, the spatiotemporal consistency score of each path is determined, and the dominant transmission path with a spatiotemporal consistency score exceeding the preset path screening threshold is selected. The pollutant is traced backward from the receptor point to the source along the dominant transport path. The starting position of the dominant transport path is identified as the spatial coordinate information of the pollution source. The mass transport flux information of each medium node passed through by the dominant transport path is extracted. The contribution of each medium to the pollution source is calculated according to the proportion of the cumulative mass of pollutants transported from each medium node to the pollution source location to the total transport mass. The medium contribution weight is generated. The spatial coordinate information and the medium contribution weight are combined to form the pollution source tracing feature set.

[0029] After the quality exchange network is established, two types of core quantitative indicators are calculated for each path of the heterogeneous environmental data collected in the target ecological area—including pollutant concentration sequences at soil sampling points, surface water and groundwater monitoring station data, atmospheric diffusion monitoring grid data, and bioaccumulation detection data—to support the subsequent selection of dominant transmission paths.

[0030] For each path in the quality exchange network, pollutant concentration observations are sequentially read from each node along the path direction. The concentration difference between adjacent nodes is divided by the spatial distance between the two nodes to obtain the local concentration gradient of each segment. All local concentration gradients are then summed along the path direction, and finally divided by the total path length to obtain the concentration gradient intensity index for that path. Let the total path length be L, and there be n nodes on the path. The pollutant concentration difference between the k-th node and the (k+1)-th node is... The length of the local segment between the two nodes is The concentration gradient intensity index G is defined as follows: This indicator reflects the strength of the average concentration driving force for pollutant propagation along the path: the higher the G value, the steeper the concentration drop along the path, the stronger the driving force for pollutant diffusion along the path, and the closer the correlation between the path and the pollution source.

[0031] The time propagation rate index is calculated as follows: Peak values ​​of pollutant concentrations at each node along the path are detected, and the moment when the concentration peak occurs at that node is extracted. The peak moment of each node is subtracted from the peak moment of the starting node to obtain the time delay of each node relative to the starting point. This time delay is then divided by the cumulative path length from the starting point to obtain the local time propagation rate at each node. Finally, the average of the local time propagation rates of all intermediate nodes along the path is taken as the time propagation rate index for that path. Let the time difference between the peak moment of the k-th node and the starting point be... The cumulative path length from this node to the starting point of the path is If the total number of intermediate nodes on the path is m, then the time propagation rate index V is defined as follows: A smaller V value means that pollutants travel faster along the path, and the ratio of time response to spatial distance is low, indicating that the path is an efficient transport channel.

[0032] Based on the above two types of indicators, the spatiotemporal consistency score S is calculated for each path. The spatiotemporal consistency score comprehensively reflects the degree of synergy between the concentration gradient driving force and the time propagation efficiency, specifically through the analysis of... and We obtain the following by weighting and combining the reciprocals: ,in and These are the dimensionless results after normalizing the G and V values ​​for all paths, respectively. α and β are weighting coefficients, satisfying α+β=1. The default values ​​are α=0.5 and β=0.5, which can be adjusted according to the media transmission characteristics of the actual ecological region. The spatiotemporal consistency score s of each path is compared with the preset path selection threshold. Compare and filter those that meet the requirements. The set of paths serves as the dominant transmission path. The value can be statistically calibrated based on historical pollution event data of the target ecological area. In the absence of historical data, the 75th percentile of all path scores can be taken as the default threshold to retain the highest-scoring quarter of the paths for subsequent source tracing analysis.

[0033] After determining the set of dominant transmission paths, reverse tracing is performed along each dominant transmission path. The direction of reverse tracing is opposite to the actual diffusion direction of pollutants: starting from the receptor point (i.e., the terminal node where pollutant concentration exceeds the standard), tracing back node by node along the path until reaching the starting position of the path. The starting position is identified as the spatial coordinate information of the potential pollution source corresponding to that path, recorded in the form of a triplet of latitude, longitude, and elevation values, forming a set of geographic coordinates of the candidate pollution source location. When the reverse tracing starting points of multiple dominant transmission paths are highly clustered in space, the weighted centroid coordinates of the cluster center are used as the spatial coordinates of the final pollution source, with the weight taken as the spatiotemporal consistency score S of each path.

[0034] During the reverse tracing process, mass transfer flux information of the media nodes traversed by each dominant transmission path is extracted simultaneously. This mass transfer flux information originates from the mass exchange rate parameters at each phase interface in the mass exchange network, characterizing the mass of contaminants transferred from one medium to an adjacent medium per unit time through that interface. For each media node on the path, the cumulative mass of contaminants transferred towards the pollution source location within the tracing time window is statistically analyzed. The subscript i represents the number of the i-th medium node. The total transmission quality is obtained by summing the cumulative transmission quality of all medium nodes. ,in This represents the total number of media nodes on the path. The media contribution weight of the i-th media node. Defined as ,satisfy This weighting system quantitatively describes the relative contributions of various media, such as soil, water, atmosphere, and organisms, to the process of pollutant convergence towards the pollution source, providing data support for the hierarchical visualization of 3D virtual labeled objects in subsequent AR scenes.

[0035] When multiple dominant transmission paths exist, the contribution weights of nodes of the same type on each path are weighted and averaged. The weight is taken as the spatiotemporal consistency score S of the corresponding path, thus obtaining the comprehensive contribution weight vector of each medium at the target ecological region level. Finally, the spatial coordinate information of the pollution source is combined with the contribution weight vector of each medium to form a pollution source tracing feature set. This feature set is stored in a structured data format. Each record contains the three-dimensional geographic coordinates of the pollution source candidate location, the contribution weight value of each medium, the corresponding dominant transmission path number, and the spatiotemporal consistency score, providing complete input data for subsequent multi-scale registration and three-dimensional virtual annotation object generation on AR devices.

[0036] In one optional implementation, based on the spatial coordinate information, extracting the edge and texture features of ground objects in the real-time field of view of the AR device, and performing multi-scale registration with the corresponding ground object features in the satellite remote sensing image includes: Based on the spatial coordinate information, the geographical location of the pollution source in the world coordinate system is determined, the field of view of the AR device is obtained, and the edge features and texture features of ground objects within the field of view are extracted. Remote sensing images covering the geographical location are retrieved from the satellite remote sensing image database. Corresponding ground features in the remote sensing images corresponding to the field of view are extracted. By constructing a multi-scale feature pyramid, the edge and texture features of ground features within the field of view are registered layer by layer with the corresponding ground features in the remote sensing images at different spatial resolution levels. At the coarsest resolution level of the multi-scale feature pyramid, the spatial position offset and rotation angle offset between the edge features of ground objects within the field of view and the corresponding edge features of ground objects in the remote sensing image are calculated. The spatial position offset and the rotation angle offset are used as initial registration parameters and passed to the next resolution level. At the next resolution level, the local deformation parameters between texture features are further calculated based on the initial registration parameters. The registration parameters are passed layer by layer until the finest resolution level, resulting in multi-level registration parameters from the world coordinate system to the device field of view coordinate system.

[0037] Based on the spatial coordinate information carried in the pollution source tracing feature set, the geographical location of the pollution source can be determined in the world coordinate system. This geographical location is expressed in the form of latitude, longitude, and elevation values, forming the benchmark reference point for subsequent AR field of view registration. The current attitude parameters of the AR device (including the device's pitch angle, yaw angle, and roll angle) and the data from the device's built-in inertial measurement unit are acquired. Combined with the device's focal length and field of view parameters, the bounding box of the geographical area covered by the current field of view is calculated. Based on this, ground feature extraction is performed on real-time image frames within the field of view, specifically including edge features and texture features. Edge features are obtained by applying multi-directional gradient operators to the image frames, and the extraction results are saved in the form of edge intensity maps and edge direction maps. Texture features are encoded by encoding image patches using local binary mode descriptors or gradient direction histogram descriptors, forming a dense texture feature description vector field that covers all ground feature areas within the field of view.

[0038] When retrieving remote sensing images covering the aforementioned geographical locations from the satellite remote sensing image database, it is necessary to filter remote sensing image tiles with the most recent timestamps and cloud cover below a preset threshold based on the latitude and longitude range of the pollution source's geographical location and the geographical boundary of the current field of view. The retrieved remote sensing images undergo radiometric calibration and atmospheric correction preprocessing, converting the raw digital quantization values ​​into surface reflectance to eliminate interference from atmospheric scattering and sensor response differences during imaging on feature extraction. Subsequently, corresponding ground feature features, including edge and texture features in the remote sensing image, are extracted from the area corresponding to the field of view on the remote sensing image. The extraction method is consistent with that of the AR device's field of view image, thus ensuring the comparability of features on both sides at the descriptor level.

[0039] The multi-scale feature pyramid is constructed as follows: The AR device's field-of-view image and the remote sensing image are downsampled to generate a multi-layered image pyramid with progressively decreasing resolution from the original. The downsampling ratio between adjacent layers is set to 2. The number of pyramid layers is obtained by rounding up the logarithm (base 2) of the ratio of the original image resolution to the target resolution of the coarsest layer. At each layer, edge and texture features at the corresponding resolution are re-extracted from both the field-of-view image and the remote sensing image, forming two sets of feature pyramids. The coarsest resolution layer has the smallest image size and contains the fewest high-frequency details, but has the most significant global structural information. It is suitable for coarse registration at this layer to quickly narrow the search space and obtain a stable initial transform estimate.

[0040] At the coarsest resolution level, the spatial position offset and rotation angle offset between the edge features of ground objects on the AR device's field-of-view image side and the corresponding edge features on the remote sensing image side are calculated. Specifically, phase correlation operations are performed on the edge feature maps of both sides. By calculating the cross-power spectrum of the Fourier transform results of the two edge intensity maps, the peak response of the translation in the frequency domain is obtained, thus yielding the spatial position offset, denoted as . ,in This represents the pixel offset in the horizontal direction. Represents the pixel offset in the vertical direction. Rotation angle offset. This is obtained by transforming the edge patterns on both sides to the polar-log domain and then performing phase correlation calculations again. This method is invariant to translation and can independently estimate rotation. As initial registration parameters, they are passed to the next resolution level.

[0041] At the next resolution level, the initial registration parameters Multiplying by the resolution scaling factor between adjacent layers (i.e., upsampling factor 2), it is mapped to the pixel coordinate system of the current layer as the initial transformation for the current layer registration search. Based on this, local deformation parameters between texture features are further calculated. These local deformation parameters are obtained by dividing the field of view image and remote sensing image into several local image patches at the current layer, and matching the texture feature description vectors of each image patch using the normalized cross-correlation coefficient as the matching criterion. For each image patch, its residual displacement vector under the initial transformation constraints is calculated, resulting in a dense local displacement vector field. This vector field is the local deformation parameter of the current layer, denoted as D, where each component of D represents the residual displacement vector at the center of the corresponding image patch. The updated global registration parameters of the current layer are then used... , , The local deformation parameter D is passed to the next level together as an initial estimate for more refined level registration.

[0042] The aforementioned layer-by-layer transfer process continues until the finest resolution level (i.e., the original resolution level) is reached. At the finest level, the global registration parameters have been progressively refined through multiple levels, and the residual registration error is concentrated in the local deformation parameter D. Thin-plate spline interpolation is performed on the local deformation parameters at the finest level to expand the sparse image block-level displacement vector field into a pixel-level dense deformation field, thereby achieving sub-pixel-level fine alignment between the field-of-view image and the remote sensing image. During this process, for occluded areas (such as building occlusion or tree canopy occlusion) or low-texture areas (such as water surfaces or bare soil), a confidence threshold for the local normalized cross-correlation coefficient is set. Eliminate those with a confidence level lower than The image patch matching results are optimized to avoid interference from low-quality matching points on deformation field estimation.

[0043] After completing the registration at the finest resolution level, a multi-level registration parameter set from the world coordinate system to the device field of view coordinate system is constructed by integrating the global registration parameters of each level and the dense deformation field of the finest level. This set contains three types of parameters: first, the global translation amount describing the overall rigid transformation. and rotation angle offset Secondly, there is the dense deformation field D describing local non-rigid deformation; and thirdly, there is the confidence mask accumulated during each level of registration, used to mark the spatial distribution of reliable and unreliable registration areas. Through these multi-level registration parameters, the spatial coordinate information of the pollution source tracing feature set can be accurately transformed from the world coordinate system to the AR device's field-of-view coordinate system, providing a high-precision spatial mapping foundation for the subsequent overlay display of 3D virtual labeled objects. In practical applications, the AR device continuously updates its attitude parameters during movement. Therefore, the multi-scale registration process incrementally updates each frame of image using a sliding window, utilizing the registration parameters of the previous frame as the initial value for the coarse registration of the current frame. This ensures registration accuracy while reducing the computational overhead of frame-by-frame re-registration, meeting the latency requirements of real-time AR display.

[0044] In one optional implementation, converting the pollution source tracing feature set into a three-dimensional virtual annotation object in the AR scene by constructing a hierarchical mapping relationship from the world coordinate system to the device field of view coordinate system includes: Based on the multi-level registration parameters, a hierarchical mapping relationship from the world coordinate system to the device field of view coordinate system is constructed. The hierarchical mapping relationship includes a global coordinate transformation matrix and a local deformation field. The global coordinate transformation matrix is ​​used to transform the spatial coordinate information from the world coordinate system to the camera coordinate system of the AR device. The local deformation field is used to compensate for local geometric distortions caused by differences in viewing angle and terrain undulations between the field of view image and the remote sensing image. The spatial coordinate information and the medium contribution weight in the pollution source tracing feature set are transformed to the device field of view coordinate system through the hierarchical mapping relationship. Based on the transformed coordinate information, a three-dimensional geometric model identifying the location of the pollution source is generated in the AR scene. Based on the medium contribution weight, visual encoding attributes are assigned to different parts of the 3D geometric model. The visual encoding attributes include color mapping and transparency mapping. The color mapping determines the display color of the corresponding part of the 3D geometric model according to the magnitude of the medium contribution weight. The transparency mapping determines the transparency value of the corresponding part of the 3D geometric model according to the confidence level of the medium contribution weight, thereby generating the 3D virtual annotation object.

[0045] like Figure 2 As shown, the method includes: After completing multi-level registration, a complete registration result containing global coarse registration parameters and layer-by-layer local residual displacements was obtained. Based on these multi-level registration parameters, a hierarchical mapping relationship from the world coordinate system to the device's field-of-view coordinate system was constructed. This hierarchical mapping relationship consists of two parts: a global coordinate transformation matrix and a local deformation field. The global coordinate transformation matrix is ​​responsible for handling coordinate transformations at the macroscopic scale, mapping the spatial coordinate information in the pollution source tracing feature set from the world coordinate system (i.e., the geographic coordinate system, a three-dimensional coordinate system expressed in latitude, longitude, and elevation) sequentially through the geocentric-ground-fixed coordinate system, the AR device body coordinate system, and finally to the camera coordinate system. This process involves the concatenated product of rotation matrices and translation vectors, where the rotation component is calculated by the attitude quaternions provided by the device's inertial measurement unit (IMU), and the translation component is determined by the device's position information provided by the GNSS module. The global coordinate transformation matrix uniformly transforms the three-dimensional point coordinates in the world coordinate system to the camera coordinate system, completing the rigid body transformation across coordinate systems.

[0046] The local deformation field is used to compensate for local geometric distortions caused by differences in viewing angle and terrain undulations between the field-of-view image and the remote sensing image. Since satellite remote sensing imagery uses orthophoto or near-orthophoto projection, while AR device cameras use perspective projection, significant local geometric misalignment occurs in areas with large terrain undulations such as mountains and hills. The local deformation field is stored as a dense displacement vector grid, where each node records the additional offset compensation required for the image coordinates at that location. After projecting the pollution source location from the world coordinate system to the camera image coordinate system, the local deformation field is queried at those coordinates, and deformation compensation is applied to the projection result to obtain the precise pixel location corresponding to the actual field-of-view image. The spatial resolution of the local deformation field is consistent with the finest level of multi-level registration, ensuring geometric alignment accuracy in detailed areas.

[0047] After transforming the spatial coordinate information and media contribution weights from the pollution source tracing feature set to the device's field-of-view coordinate system through the aforementioned hierarchical mapping relationship, a 3D geometric model identifying the pollution source location is generated in the AR scene. The spatial shape of the 3D geometric model is determined based on the pollution source's influence range described by the pollution source tracing feature set: if the tracing result indicates a point pollution source, a 3D sphere or cone centered on that point is generated; if the tracing result indicates a planar or linear pollution source (such as a sewage ditch or farmland non-point source), a corresponding 3D cylinder, patch, or tubular body extending along the path is generated. The size parameters of the 3D geometric model are set based on the spatial uncertainty estimate in the pollution source tracing feature set; the greater the uncertainty, the larger the model's outer envelope size, to intuitively reflect the positioning accuracy of the tracing result. The 3D geometric model is presented on the real-time field of view in the AR scene through overlay rendering, maintaining spatial alignment with real-world features.

[0048] Visual encoding attributes are assigned to different parts of the 3D geometric model based on the medium's contribution weight. These visual encoding attributes include two dimensions: color mapping and transparency mapping. The color mapping maps the numerical value of the medium's contribution weight to a predefined color gradient scheme. Specifically, let the contribution weight corresponding to the j-th medium category be... ,Will After normalization to the [0, 1] interval, the display color of the corresponding part in the 3D geometric model is determined according to a continuous color gradient scheme from cool tones (blue-green, corresponding to low contribution weight) to warm tones (orange-red, corresponding to high contribution weight). When the contribution weight of a certain medium (such as surface water) is significantly higher than that of other media, its corresponding model area is displayed in dark red, prompting on-site personnel to pay close attention to the sampling and verification of this medium; when the contribution weight is low, the corresponding area is displayed in blue-green, indicating that the contribution of this medium to pollution transmission is relatively limited. The color mapping scheme is simultaneously displayed in the legend area of ​​the AR display interface so that users can intuitively read the correspondence between color and weight values.

[0049] Transparency mapping determines the transparency value of the corresponding part of the 3D geometric model based on the credibility of the medium contribution weight. The credibility quantification is derived from the confidence interval estimation of the mass exchange calculation for each medium node during the pollution source tracing process: if the monitoring data of a certain medium node within the tracing time window is relatively complete and the concentration time-series fluctuations are highly regular, then the credibility of the node's contribution weight is high, and the transparency of the corresponding model part is low (i.e., high opacity), appearing as a clearly visible entity in the AR field of view; conversely, if the monitoring data of a certain medium node has many missing or outlier values, resulting in greater uncertainty in the contribution weight, then the transparency of the corresponding model part is high, appearing as semi-transparent or even nearly transparent in the AR field of view, indicating to the user that the tracing conclusion for this area needs further verification. Let the credibility score of the contribution weight of the j-th medium category be . (The value ranges from [0, 1], with larger values ​​indicating higher reliability), then the corresponding opacity according to The linear mapping method is determined, where and These are the preset minimum and maximum opacity boundary values, typically set to 0.2 and 1.0, to prevent the model from completely disappearing and causing information loss.

[0050] The combined effect of color mapping and transparency mapping constitutes the final visual presentation of the 3D virtual annotation object. The 3D virtual annotation object observed by the user in the AR device's field of view simultaneously conveys two types of information: color reflects the relative contribution of each medium to pollution transmission, and transparency reflects the credibility of this judgment. This dual-channel visual encoding method allows field personnel to directly obtain key information for pollution source tracing from the AR field of view without consulting digital reports, quickly determining which areas require priority on-site sampling verification and which medium pathway tracing conclusions have a high degree of confidence. The 3D virtual annotation object also includes an interactive interface, allowing users to expand a detailed numerical information panel through touch or gesture operations to view the specific values ​​and confidence intervals of the contribution weights of each medium, thereby supporting subsequent selection of on-site verification points and actual data collection.

[0051] In one optional implementation, receiving the on-site verification point selected by the user through the interactive interface of the three-dimensional virtual annotation object and the corresponding measured pollutant concentration value, and constructing the residual distribution field between the measured pollutant concentration value and the predicted concentration value of the pollution source tracing feature set includes: The system receives touch operation commands from the user on the interactive interface of the three-dimensional virtual annotation object, and marks the three-dimensional spatial position of the on-site verification point in the AR scene according to the touch operation commands. The three-dimensional spatial position is converted into geographic coordinates in the world coordinate system through the inverse mapping of the hierarchical mapping relationship. Receive the measured pollutant concentration values ​​collected by the user at the on-site verification point; The predicted concentration values ​​at the corresponding geographical coordinates of the field verification points are extracted from the pollution source tracing feature set. The predicted concentration values ​​are calculated by the mass exchange network based on the interphase migration characteristics and the medium contribution weight. For each medium at the field verification point, calculate the concentration residual value between the measured pollutant concentration value and the predicted concentration value. When there are multiple field verification points, summarize the concentration residual values ​​of all field verification points to form a residual dataset. Using the geographic coordinates of each field verification point in the residual dataset as spatial anchors, the discrete concentration residual values ​​are extended into a continuously distributed residual distribution field within the target ecological area through spatial interpolation. The spatial interpolation method calculates the residual estimate of the unsampled location based on the distance decay principle. The residual estimate gradually decays to zero as the distance between it and the nearest field verification point increases.

[0052] On the AR device's display interface, 3D virtual annotation objects are presented to the user as overlays on the real scene. Users can perform operations such as clicking, long-pressing, and dragging these virtual annotation objects through the touch screen. When a user triggers a selection operation on the interactive interface, the touch operation command carries the screen's 2D coordinate information. This coordinate information undergoes coordinate system transformation through inverse mapping of the layered mapping relationship. Specifically, the inverse mapping process first restores the screen coordinates from the device's field of view coordinate system to the camera coordinate system. Then, combined with the device's real-time pose parameters (including rotation matrices and translation vectors), the ray direction in the camera coordinate system is transformed to the world coordinate system. Subsequently, by intersecting with the digital elevation model or surface geometry model, the 3D intersection point of the ray and the surface is determined, thus obtaining the geographic coordinates of the on-site verification point in the world coordinate system. These geographic coordinates are recorded in the form of latitude, longitude, and altitude, and stored together with the corresponding verification point identifier for subsequent residual calculations.

[0053] After arriving at the on-site verification point, users collect samples from various media (soil, water, atmosphere, organisms, etc.) or use portable sensors for in-situ detection to obtain the measured concentration values ​​of the target pollutants in each media. These measured concentration values ​​are submitted through the data entry function of the AR interactive interface, supporting both manual keyboard input and direct transmission with external sensor devices via Bluetooth / wired interface. Each measured data record includes the geographical coordinates of the verification point, media category identifier, pollutant type identifier, and measured concentration value, forming a structured measured data entry.

[0054] When extracting predicted concentration values ​​corresponding to the geographic coordinates of field verification points from the source tracing feature set, if the geographic coordinates of the verification point coincide with a certain grid node in the source tracing feature set, the predicted concentration value of that node is directly read; if the verification point falls between grid nodes, bilinear interpolation is performed on the predicted concentration values ​​of the surrounding neighboring grid nodes to obtain the estimated predicted concentration value for that location. The predicted concentration value is calculated by the mass exchange network based on the interphase migration characteristics and media contribution weights of each medium, reflecting the model's theoretical expectation of the pollutant concentration of each medium at that location.

[0055] For each field verification point, the difference between the measured pollutant concentration and the predicted concentration is calculated according to the media category; this is the concentration residual. Let the measured concentration at a certain verification point for the p-th media be... The corresponding predicted concentration value is Then the concentration residual value of the medium for: ; when When this occurs, it indicates that the model underestimates the pollutant concentration of the medium at that location; when When the value is too high, it indicates that the model overestimates. The concentration residuals for all field validation points and all media categories are summarized to form a residual dataset. Each record in the residual dataset contains three core fields: geographic coordinates, media category, and residual value. Together, they form a discrete set of residual samples distributed within the target ecological area.

[0056] To extend the discrete concentration residuals to a continuous residual distribution field within the target ecological area, a spatial interpolation method based on the distance decay principle is adopted. The core idea of ​​this method is that the closer an unsampled location is to a known verification point, the greater the influence of that verification point on its residual estimate; as the distance increases, the influence gradually decreases, and when the distance exceeds a certain range, the residual estimate approaches zero, indicating that the location lacks reliable measured constraints.

[0057] For any unsampled location within the target ecological area, let its geographical coordinates be... The residual dataset contains Q field verification points, and the geographical coordinates of the q-th verification point are: The concentration residual value corresponding to medium p is .definition and The spatial distance between them is Then the position Residual estimate of medium p Calculated using inverse distance weighted interpolation: ; in The distance decay weight function is defined as follows: , This is the distance decay index, which controls the rate at which the weight decays with distance. It typically ranges from 1 to 3 and can be adjusted based on the pollutant diffusion characteristics of the target ecological area. With a certain verification point Overlap (i.e.) When ), directly order This avoids abnormal values ​​where the denominator is zero.

[0058] To constrain the residual estimate to decay to zero with increasing distance, a cutoff radius is introduced. ,when season This means that verification points outside the cutoff radius are not included in the residual estimation for that location. If the location... There are no verification points within the cutoff radius (i.e., all) ), then let This indicates that the residual estimate at that location is zero, and the model prediction is considered unbiased at that point. Cutoff radius The value is determined based on the area of ​​the target ecological region and the spatial distribution density of the verification points, ensuring that the residual distribution field smoothly transitions to zero in the sparse data region and avoids the accumulation of extrapolation errors.

[0059] The above interpolation calculations are performed on all grid locations within the target ecological area to generate continuous residual distribution field raster data for each medium. Each medium corresponds to one layer of residual distribution field, and multiple layers of residual distribution fields together constitute a multi-medium residual distribution field set, providing complete spatial error information for subsequent identification of spatial anomaly areas and adjustment of model parameters. The spatial resolution of the residual distribution field is consistent with the grid resolution of the pollution source tracing feature set, facilitating comparative analysis and iterative updates of both within the same coordinate framework.

[0060] On the AR display interface, the residual distribution field is superimposed on the three-dimensional virtual scene in the form of a color-coded heat map. Positive residual regions (model underestimation regions) and negative residual regions (model overestimation regions) are presented with different hues, allowing users to intuitively perceive the spatial distribution pattern of model prediction errors. This helps to identify spatially abnormal regions that need to be focused on later, and provides a visual basis for adjusting the medium contribution weight and interphase migration characteristics.

[0061] In one optional implementation, spatially anomalous regions in the residual distribution field are identified, and the medium contribution weight and the interphase migration characteristics are adjusted according to the distribution characteristics of the spatially anomalous regions. The pollution source tracing feature set is then recalculated to obtain the ecological pollution source location results, including: Calculate the global residual statistical characteristics of the residual distribution field; Based on the global residual statistical characteristics, a residual anomaly determination threshold is determined. Grid nodes in the residual distribution field whose absolute value of the residual estimate exceeds the residual anomaly determination threshold are marked as abnormal nodes. Connected regions formed by spatially adjacent abnormal nodes are identified as spatial abnormal regions. The geometric center location and residual amplitude characteristics of the spatial anomaly region are extracted as distribution features; The geometric center of the spatial anomaly region is mapped to the node space of the quality exchange network, the medium type corresponding to the geometric center is identified, the adjustment direction of the medium contribution weight is determined according to the positive or negative sign of the residual amplitude feature, and the adjustment magnitude of the medium contribution weight is determined according to the average amplitude of the residual amplitude feature. The phase migration characteristics between the medium type and adjacent medium types are adjusted according to the coverage area and boundary shape of the spatial anomaly region. The adjusted medium contribution weight and the adjusted interphase migration characteristics are input into the mass exchange network, and the reverse tracing calculation of the transmission path from the receptor point to the source is re-executed to obtain the updated pollution source tracing feature set. The spatial coordinate information in the updated pollution source tracing feature set is used as the ecological pollution source location result.

[0062] After obtaining the residual distribution field, statistical analysis is required to identify spatial regions where model prediction biases are concentrated. The mean value is calculated for the residual estimates of all grid nodes in the residual distribution field. and standard deviation This serves as a statistical feature of the global residuals. Based on the normal distribution assumption, the residual anomaly determination threshold is set... Set as ,in This is a multiplier, typically between 2 and 3, with the specific value adjusted based on the pollution complexity of the target ecological area. For residual estimates whose absolute value exceeds... The grid nodes are marked as anomalous nodes. After marking the anomalous nodes, a connected component analysis method is used to cluster spatially adjacent anomalous nodes. Sets of anomalous nodes that are spatially consecutive (including four-neighbor or eight-neighbor connectivity) are identified as a connected region, i.e., a spatial anomalous region. Connected regions with too small an area (number of nodes below a preset minimum size threshold) are considered noise and filtered out, and not included in subsequent analysis.

[0063] For each identified spatial anomaly region, its geometric center location and residual amplitude features are extracted as distribution features. The geometric center location is obtained by calculating the mean geographic coordinates of all anomaly nodes within the connected region, thus yielding the center latitude and longitude coordinates of the region. The residual magnitude characteristics include two aspects: first, the sign distribution of the residual estimates for all outlier nodes within the region, i.e., the ratio of positive to negative residual nodes; and second, the average magnitude of the absolute values ​​of the residual estimates within the region. The residuals reflect the degree of model bias in that region. Positive residuals indicate that the model-predicted concentration is lower than the measured concentration, suggesting that the contribution of pollutant sources in that region is underestimated; negative residuals indicate that the model-predicted concentration is higher than the measured concentration, suggesting that the contribution of pollutant sources in that region is overestimated.

[0064] The geometric center of the spatial anomaly region Mapped to the node space of the quality exchange network, the spatial distance between this coordinate and the geographical coordinates of each media node in the network is calculated to find the network node closest to the geometric center, identifying its corresponding media type, such as soil, water, atmosphere, or biological media. The direction of adjustment for the media contribution weight is determined based on the positive or negative sign of the residual amplitude characteristics: if the spatial anomaly region is dominated by positive residuals, the contribution weight of the corresponding media needs to be increased; if it is dominated by negative residuals, the contribution weight of the corresponding media needs to be decreased. The adjustment magnitude is based on the average amplitude. Specifically, will The predicted average concentration of this medium in the current source tracing feature set The ratio is used as the relative adjustment ratio. ,Right now Based on this, the contribution weight of the medium node is positively or negatively corrected. The corrected weight needs to be normalized to ensure that the sum of the contribution weights of all medium nodes remains at 1.

[0065] Based on the coverage area and boundary shape of the spatial anomaly region, the interphase migration characteristics between the corresponding medium type and adjacent medium types are adjusted. A large coverage area of ​​the spatial anomaly region indicates that the spatial range of pollutant cross-media transport is underestimated or overestimated, requiring an increase or decrease in the effective mass transfer area parameter of the corresponding phase interface. The boundary shape reflects the directional characteristics of pollutant migration. If the boundary shape exhibits obvious strip-like or directional extension, it indicates that the pollutant migration flux in that direction is strong, requiring a corresponding increase in the interphase migration coefficient in that direction. If the boundary shape is relatively circular or has no obvious directionality, the interphase migration characteristics are adjusted proportionally in an isotropic manner. The adjustment amount of the interphase migration characteristics is determined by the area of ​​the spatial anomaly region. Compared with reference area The ratio of (i.e., the initial effective area of ​​the interface in the quality exchange network) To quantify this, the interphase migration coefficient is multiplied by an adjustment factor. This enables dynamic correction of the transphase interface transport capability of pollutants.

[0066] After adjusting the media contribution weights and interphase migration characteristics, the updated parameters are re-input into the mass exchange network. Starting from the receptor point (i.e., the monitoring point with high pollutant concentration) in the target ecological area, the reverse tracing calculation of the transport path from the receptor point to the source is re-executed. During the reverse tracing process, based on the updated concentration gradient spatiotemporal correlation pattern, the pollutant transport paths between each media are re-determined, and the contributions of each path are weighted and fused according to the updated media contribution weights to obtain the updated pollution source tracing feature set. This feature set contains the corrected spatial coordinate information and the contribution weight distribution of each media, where the spatial coordinate information corresponds to the most likely original emission location or accumulation location of the pollutants.

[0067] The updated spatial coordinate information from the pollution source tracing feature set serves as the final output for locating the ecological pollution source. This coordinate information can be directly overlaid onto a 3D virtual annotation object in the AR scene, displaying the precise geographical location of the pollution source as a highlighted marker in the device's field of view, allowing on-site personnel to further confirm and address the issue. In practical applications, the aforementioned residual analysis and weight adjustment process can be iteratively executed. After each iteration, the measured data from the verification points are re-collected, and the residual distribution field is updated until the spatial anomaly areas in the residual distribution field disappear or the residual amplitude drops to a preset convergence threshold. The iteration then terminates, and the final ecological pollution source location result is output. The entire closed-loop correction mechanism effectively improves the spatial accuracy of pollution source location, especially in ecological scenarios with multiple superimposed pollution sources or complex migration paths between media.

[0068] A second aspect of the present invention provides an AR-assisted ecological pollution source localization system, comprising: The pollution source tracing unit is used to construct a mass exchange network of pollutants across phase interfaces based on the interphase migration characteristics of pollutants in soil-water-atmosphere-biological multi-media and the spatiotemporal correlation patterns of pollutant concentration gradients between different media. The mass exchange network traces the transmission path of pollutants from the receptor point to the source by identifying the spatiotemporal correlation patterns of pollutant concentration gradients in different media in the heterogeneous environmental data of the target ecological area, and obtains a pollution source tracing feature set containing spatial coordinate information and media contribution weights. The processing unit is used to extract the edge features and texture features of ground objects in the real-time field of view of the AR device based on the spatial coordinate information, perform multi-scale registration with the corresponding ground object features in the satellite remote sensing image, and convert the pollution source tracing feature set into a three-dimensional virtual annotation object in the AR scene by constructing a hierarchical mapping relationship from the world coordinate system to the device field of view coordinate system. The processing unit is used to receive the on-site verification points and corresponding measured pollutant concentration values ​​selected by the user through the interactive interface of the three-dimensional virtual annotation object, construct the residual distribution field between the measured pollutant concentration values ​​and the predicted concentration values ​​of the pollution source tracing feature set, identify spatially anomalous regions in the residual distribution field, adjust the medium contribution weight and the interphase migration characteristics according to the distribution characteristics of the spatially anomalous regions, recalculate the pollution source tracing feature set, and obtain the ecological pollution source location result.

[0069] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0070] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0071] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AR-assisted method for locating ecological pollution sources, characterized in that, include: Based on the interphase migration characteristics of pollutants in multiple media (soil, water, atmosphere, and organisms) and the spatiotemporal correlation patterns of pollutant concentration gradients between different media in heterogeneous environmental data of the target ecological region, a mass exchange network of pollutants across phase interfaces is constructed, including: By extracting soil microbial community structure data from the heterogeneous environmental data, the spatial distribution gradient of specific degrading bacteria was analyzed, and the spatial coupling relationship between changes in degrading bacteria density and changes in pollutant concentration was identified. Extract vegetation spectral response data from the heterogeneous environmental data, identify the spatial aggregation pattern and spectral reflectance shift amplitude of chlorophyll fluorescence anomaly regions, and determine the accumulation path and residence time characteristics of pollutants in biological media based on the spatial coupling relationship, the spatial aggregation pattern and the spectral reflectance shift amplitude. The accumulation path and residence time characteristics are cross-validated with the spatiotemporal distribution data of surface pollutant concentration and groundwater flow vector field data in the heterogeneous environmental data. By comparing the time series of pollutant accumulation in biological media with the time series of pollutant migration in non-biological media, the spatiotemporal correlation patterns of pollutant concentration gradients between different media are identified. Based on the spatiotemporal correlation pattern and the interphase migration characteristics of pollutants in soil-water-atmosphere-biological multi-media, a mass exchange network spanning abiotic and biological media is constructed. The mass exchange network identifies the spatiotemporal correlation pattern of pollutant concentration gradients in different media in heterogeneous environmental data of the target ecological region, and traces the transmission path of pollutants from the receptor point to the source in reverse, obtaining a pollution source tracing feature set containing spatial coordinate information and media contribution weights. Based on the spatial coordinate information, the edge features and texture features of ground objects in the real-time field of view image of the AR device are extracted and registered with the corresponding ground object features in the satellite remote sensing image at multiple scales. By constructing a hierarchical mapping relationship from the world coordinate system to the device field of view coordinate system, the pollution source tracing feature set is converted into a three-dimensional virtual annotation object in the AR scene. The system receives the on-site verification points and corresponding measured pollutant concentration values ​​selected by the user through the interactive interface of the three-dimensional virtual annotation object. It constructs a residual distribution field between the measured pollutant concentration values ​​and the predicted concentration values ​​of the pollution source tracing feature set, identifies spatially anomalous regions in the residual distribution field, adjusts the medium contribution weight and the interphase migration characteristics according to the distribution characteristics of the spatially anomalous regions, recalculates the pollution source tracing feature set, and obtains the ecological pollution source location result.

2. The method according to claim 1, characterized in that, The quality exchange network identifies the spatiotemporal correlation patterns of pollutant concentration gradients in different media within heterogeneous environmental data of the target ecological region, and traces the transport path of pollutants from receptor points to sources in reverse, obtaining a pollution source tracing feature set containing spatial coordinate information and media contribution weights, including: The ratio of the cumulative value of pollutant concentration gradient along the path direction to the path length is calculated as a concentration gradient intensity index, and the ratio of the time difference between the peak pollutant concentration along the path direction and the path starting time to the path length is calculated as a time propagation rate index. Based on the concentration gradient intensity index and the time propagation rate index, the spatiotemporal consistency score of each path is determined, and the dominant transmission path with a spatiotemporal consistency score exceeding the preset path screening threshold is selected. The pollutant is traced backward from the receptor point to the source along the dominant transport path. The starting position of the dominant transport path is identified as the spatial coordinate information of the pollution source. The mass transport flux information of each medium node passed through by the dominant transport path is extracted. The contribution of each medium to the pollution source is calculated according to the proportion of the cumulative mass of pollutants transported from each medium node to the pollution source location to the total transport mass. The medium contribution weight is generated. The spatial coordinate information and the medium contribution weight are combined to form the pollution source tracing feature set.

3. The method according to claim 1, characterized in that, Based on the spatial coordinate information, the edge and texture features of ground objects in the real-time field-of-view image of the AR device are extracted, and multi-scale registration is performed with the corresponding ground object features in the satellite remote sensing image, including: Based on the spatial coordinate information, the geographical location of the pollution source in the world coordinate system is determined, the field of view of the AR device is obtained, and the edge features and texture features of ground objects within the field of view are extracted. Remote sensing images covering the geographical location are retrieved from the satellite remote sensing image database. Corresponding ground features in the remote sensing images corresponding to the field of view are extracted. By constructing a multi-scale feature pyramid, the edge and texture features of ground features within the field of view are registered layer by layer with the corresponding ground features in the remote sensing images at different spatial resolution levels. At the coarsest resolution level of the multi-scale feature pyramid, the spatial position offset and rotation angle offset between the edge features of ground objects within the field of view and the corresponding edge features of ground objects in the remote sensing image are calculated. The spatial position offset and the rotation angle offset are used as initial registration parameters and passed to the next resolution level. At the next resolution level, the local deformation parameters between texture features are further calculated based on the initial registration parameters. The registration parameters are passed layer by layer until the finest resolution level, resulting in multi-level registration parameters from the world coordinate system to the device field of view coordinate system.

4. The method according to claim 3, characterized in that, By constructing a hierarchical mapping relationship from the world coordinate system to the device field-of-view coordinate system, the pollution source tracing feature set is converted into a three-dimensional virtual annotation object in the AR scene, including: Based on the multi-level registration parameters, a hierarchical mapping relationship from the world coordinate system to the device field of view coordinate system is constructed. The hierarchical mapping relationship includes a global coordinate transformation matrix and a local deformation field. The global coordinate transformation matrix is ​​used to transform the spatial coordinate information from the world coordinate system to the camera coordinate system of the AR device. The local deformation field is used to compensate for local geometric distortions caused by differences in viewing angle and terrain undulations between the field of view image and the remote sensing image. The spatial coordinate information and the medium contribution weight in the pollution source tracing feature set are transformed to the device field of view coordinate system through the hierarchical mapping relationship. Based on the transformed coordinate information, a three-dimensional geometric model identifying the location of the pollution source is generated in the AR scene. Based on the medium contribution weight, visual encoding attributes are assigned to different parts of the 3D geometric model. The visual encoding attributes include color mapping and transparency mapping. The color mapping determines the display color of the corresponding part of the 3D geometric model according to the magnitude of the medium contribution weight. The transparency mapping determines the transparency value of the corresponding part of the 3D geometric model according to the confidence level of the medium contribution weight, thereby generating the 3D virtual annotation object.

5. The method according to claim 1, characterized in that, Receiving the on-site verification points and corresponding measured pollutant concentration values ​​selected by the user through the interactive interface of the 3D virtual annotation object, and constructing the residual distribution field between the measured pollutant concentration values ​​and the predicted concentration values ​​of the pollution source tracing feature set includes: The system receives touch operation commands from users on the interactive interface of the three-dimensional virtual annotation object, and marks the three-dimensional spatial position of the on-site verification point in the AR scene according to the touch operation commands. The three-dimensional spatial position is converted into geographic coordinates in the world coordinate system through the inverse mapping of the hierarchical mapping relationship. Receive the measured pollutant concentration values ​​collected by the user at the on-site verification point; The predicted concentration values ​​at the corresponding geographical coordinates of the field verification points are extracted from the pollution source tracing feature set. The predicted concentration values ​​are calculated by the mass exchange network based on the interphase migration characteristics and the medium contribution weight. For each medium at the field verification point, calculate the concentration residual value between the measured pollutant concentration value and the predicted concentration value. When there are multiple field verification points, summarize the concentration residual values ​​of all field verification points to form a residual dataset. Using the geographic coordinates of each field verification point in the residual dataset as spatial anchors, the discrete concentration residual values ​​are extended into a continuously distributed residual distribution field within the target ecological area through spatial interpolation. The spatial interpolation method calculates the residual estimate of the unsampled location based on the distance decay principle. The residual estimate gradually decays to zero as the distance between it and the nearest field verification point increases.

6. The method according to claim 1, characterized in that, Identify spatially anomalous regions in the residual distribution field, and adjust the medium contribution weight and interphase migration characteristics based on the distribution characteristics of the spatially anomalous regions. Recalculate the pollution source tracing feature set to obtain the ecological pollution source location results, including: Calculate the global residual statistical characteristics of the residual distribution field; Based on the global residual statistical characteristics, a residual anomaly determination threshold is determined. Grid nodes in the residual distribution field whose absolute value of the residual estimate exceeds the residual anomaly determination threshold are marked as abnormal nodes. Connected regions formed by spatially adjacent abnormal nodes are identified as spatial abnormal regions. The geometric center location and residual amplitude characteristics of the spatial anomaly region are extracted as distribution features; The geometric center of the spatial anomaly region is mapped to the node space of the quality exchange network, the medium type corresponding to the geometric center is identified, the adjustment direction of the medium contribution weight is determined according to the positive or negative sign of the residual amplitude feature, and the adjustment magnitude of the medium contribution weight is determined according to the average amplitude of the residual amplitude feature. The phase migration characteristics between the medium type and adjacent medium types are adjusted according to the coverage area and boundary shape of the spatial anomaly region. The adjusted medium contribution weight and the adjusted interphase migration characteristics are input into the mass exchange network, and the reverse tracing calculation of the transmission path from the receptor point to the source is re-executed to obtain the updated pollution source tracing feature set. The spatial coordinate information in the updated pollution source tracing feature set is used as the ecological pollution source location result.

7. An AR-assisted ecological pollution source localization system, used to implement the method as described in any one of claims 1-6, characterized in that, include: The pollution source tracing unit is used to construct a mass exchange network of pollutants across phase interfaces based on the interphase migration characteristics of pollutants in soil-water-atmosphere-biological multi-media and the spatiotemporal correlation patterns of pollutant concentration gradients between different media in the heterogeneous environmental data of the target ecological area. The mass exchange network traces the transport path of pollutants from the receptor point to the source by identifying the spatiotemporal correlation patterns of pollutant concentration gradients between different media in the heterogeneous environmental data of the target ecological area, and obtains a pollution source tracing feature set containing spatial coordinate information and media contribution weights. The processing unit is used to extract the edge features and texture features of ground objects in the real-time field-of-view image of the AR device based on the spatial coordinate information, perform multi-scale registration with the corresponding ground object features in the satellite remote sensing image, and convert the pollution source tracing feature set into a three-dimensional virtual annotation object in the AR scene by constructing a hierarchical mapping relationship from the world coordinate system to the device field-of-view coordinate system. The processing unit is used to receive the on-site verification points and corresponding measured pollutant concentration values ​​selected by the user through the interactive interface of the three-dimensional virtual annotation object, construct the residual distribution field between the measured pollutant concentration values ​​and the predicted concentration values ​​of the pollution source tracing feature set, identify spatially anomalous regions in the residual distribution field, adjust the medium contribution weight and the interphase migration characteristics according to the distribution characteristics of the spatially anomalous regions, recalculate the pollution source tracing feature set, and obtain the ecological pollution source location result.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images