A Method for Risk Assessment of River Sand Mining Anomalies Based on Multi-Source Monitoring Data

Abstract

This invention discloses a method for risk assessment of river sand mining anomalies based on multi-source monitoring data. The method extracts river anomaly features from multi-temporal satellite remote sensing data, UAV remote sensing data, and spatially constrained data. It then normalizes these features to calculate a comprehensive anomaly index, filters suspected anomaly candidate areas based on the index, and conducts risk assessments on these areas. After UAV inspection and verification of these candidate areas, high-risk areas confirmed by the UAV undergo 3D verification. Finally, the results are output and the data archive is updated. This invention achieves screening of suspected anomaly areas in rivers, risk classification, UAV verification, 3D evidence chain verification, and dynamic updating of inspection archives. This improves the timeliness of detecting suspected illegal sand mining anomalies in rivers, thus providing more efficient and reliable technical support for river sand mining supervision.

Description

Technical Field

[0001] This invention relates to the field of river sand mining risk assessment technology, specifically a method for assessing abnormal risks in river sand mining based on multi-source monitoring data. Background Technology

[0002] River sand mining activities are closely related to riverbed evolution, shoreline stability, flood control safety, and aquatic ecological environment protection. With the increasing requirements for river management and ecological environment protection, the demand for monitoring abnormal sand mining activities in river channels, water areas, tidal flats, and shorelines is growing.

[0003] With the development of remote sensing monitoring technology, satellite remote sensing, UAV remote sensing, and intelligent identification technology are gradually being applied to river inspection scenarios. However, existing remote sensing monitoring technologies still have the following shortcomings in the application of abnormal sand mining in river inspections: First, while satellite remote sensing is suitable for large-scale anomaly screening, it is easily affected by factors such as imaging conditions, cloud cover, temporal resolution, spectral similarity of ground objects, and natural erosion and deposition. Its ability to distinguish between human disturbances and natural topographic changes is limited, leading to insufficient stability in identifying suspected anomaly areas. Meanwhile, although drone inspections can provide high-resolution imagery or point cloud data, the lack of front-end anomaly screening, risk scoring, and task triggering mechanisms can easily result in unreasonable inspection path planning, low response efficiency, and insufficient utilization of equipment resources.

[0004] Second, although some existing solutions introduce multi-source data for joint analysis, they are mostly limited to result superposition, single comparison or single-level identification. They have not yet formed a closed-loop processing flow from wide-area anomaly screening, risk classification, rapid drone verification, three-dimensional quantitative verification to result output, and also lack a structured result expression method for regulatory applications. Summary of the Invention

[0005] The purpose of this invention is to provide a method for assessing the risk of abnormal river sand mining based on multi-source monitoring data. By collaboratively analyzing multi-temporal optical satellite imagery and synthetic aperture radar imagery, multiple features characterizing abnormal changes in the river channel are extracted and a comprehensive anomaly index is constructed. Based on this, risk scoring and classification are performed by combining spatial constraints and historical anomaly distribution information. When the risk score reaches a preset threshold, a rapid UAV verification task is triggered. Visible light imagery, thermal infrared imagery, and 3D point cloud data are combined to identify and quantify on-site targets and abnormal terrain, thereby improving the timeliness of suspected abnormal area detection, the targeting of inspection tasks, and the standardization and usability of the results output.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for assessing the risk of abnormal sand mining in river channels based on multi-source monitoring data, comprising the following steps: Acquire multi-temporal remote sensing data and auxiliary constraint data within the target area and perform preprocessing. The satellite remote sensing data includes at least one of optical satellite imagery and synthetic aperture radar (SAR) imagery; the spatial constraint data includes at least one of river management area, mining ban zone boundary, key sensitive bank section boundary, historical anomaly distribution information, hydrological information, and road accessibility information.

[0007] River channel anomaly features are extracted based on preprocessed multi-temporal remote sensing data and auxiliary constraint data. The anomaly features of each river channel are normalized to calculate the comprehensive anomaly index. Suspected anomaly candidate areas are screened based on the comprehensive anomaly index and risk assessment is conducted on the suspected anomaly candidate areas. Drones were used to inspect and verify suspected anomaly candidate areas, and the overall score of the drone verification was calculated. When the comprehensive risk score of suspected abnormal candidate areas And the overall score of drone review When the candidate region is identified as a high-risk region to be verified in three dimensions, three-dimensional verification is performed on the high-risk region to be verified in three dimensions. in, This indicates the threshold for triggering detailed 3D verification. In the absence of historical 3D verification samples for the target river section, the threshold can be determined based on the expert experience of river sand mining regulators, drone inspectors, and remote sensing interpreters. The initial value is set to 0.65. This value indicates that the UAV verification results have reached a medium-to-high level of anomaly support. After accumulating local measured data, the threshold is calibrated using historical samples that simultaneously contain UAV verification results, 3D verification results, and on-site investigation results. Specifically, based on the 3D verification or on-site investigation results, the historical samples are divided into anomaly confirmation samples and anomaly exclusion samples. By comparing the anomaly recall rate, false negative rate, and false trigger rate under different candidate thresholds, the updated threshold is determined. In regulatory scenarios where the priority is to avoid missed detections, the threshold with the lowest false trigger rate when the recall rate of confirmed abnormal samples reaches the preset requirements can be selected.

[0008] The inspection results corresponding to each candidate abnormal area are stored and updated in a structured manner to form an abnormal area inspection file.

[0009] According to the above technical solution, an adaptive weight allocation method based on softmax is used to determine the weight of each of the river channel anomaly features. The river channel anomaly features are then normalized and weighted and fused to obtain a comprehensive anomaly index. The river channel anomaly features include water boundary change index, shoreline disturbance index, exposed surface change index, vegetation change index, water turbidity change index, suspected storage yard change index, transportation channel change index, and temporary wharf change index.

[0010] According to the above technical solution, when the first [unit / item] in the target area The comprehensive anomaly index of each candidate region At that time, it was identified as a suspected anomaly candidate region; among them, The threshold for anomaly identification is set; a risk assessment is performed on the suspected anomaly candidate regions, the steps of which include: Calculate the sensitivity proximity index, historical anomaly correlation index, and whistleblower support score for suspected anomaly candidate regions; A comprehensive risk score is obtained by weighting the comprehensive anomaly index, sensitivity proximity index, historical anomaly correlation index, and the score supporting reported clues for suspected anomaly candidate regions. ; ; in, The weighting coefficients for each characteristic anomaly index are adjusted based on expert experience or historical data.

[0011] Based on comprehensive risk score Risk classification for suspected abnormal areas: ; In the formula, Indicates a low-risk area. This indicates a medium-risk area. Indicates a high-risk area. This indicates the first risk threshold. This represents the second risk threshold. The threshold... and It can be determined based on the statistical distribution of historical samples, the results of validation set optimization, or on-site inspection experience.

[0012] Candidate anomaly regions are generated based on a comprehensive anomaly index. Further analysis is conducted by combining the spatial relationship between these candidate anomaly regions and spatial constraint data, historical anomaly distribution information, and temporal sensitivity to screen and assess the risks of these regions, aiming to mitigate the impacts of natural erosion and sedimentation, seasonal water level changes, and non-sand mining activities. The risk assessment comprehensively considers the intensity of anomaly changes, spatial location characteristics, historical frequency of occurrence, proximity to sensitive areas, and the degree of support from auxiliary clues, resulting in corresponding risk scores and risk classifications. These risk classifications are used to determine the triggering conditions, inspection priorities, and rapid UAV verification range for subsequent inspection tasks.

[0013] Among these, the risk of illegal activity is typically higher if the suspected abnormal candidate area is close to the boundary of the prohibited mining area, a key monitored river section, or an ecologically sensitive area. Therefore, the shortest distance from the candidate area to the boundary of the sensitive area is set as follows: Then the sensitivity proximity index : ; in, The distance attenuation coefficient can be determined based on the target river section scale, the width of the management buffer zone, or the statistical results of historical anomaly spatial distribution. The closer the value is to 1, the closer the candidate anomaly region is to the sensitive region, and the higher its potential risk.

[0014] Sensitive area boundaries refer to spatial boundaries within the target river section that are closely related to sand mining supervision, water ecological protection, flood control safety, or shoreline stability. These boundaries include at least one of the following: no-mining zone boundaries, key monitored river section boundaries, ecologically sensitive area boundaries, important bank slope protection zone boundaries, river management area boundaries, and legal sand mining or dredging project permit boundaries.

[0015] If anomalies related to sand mining activities have occurred in the area, the probability of anomalies occurring again in the area is relatively high. Therefore, let the historical anomaly frequency of the spatial unit containing the candidate anomaly region be . The highest frequency of historical anomalies in the study area was The historical abnormal correlation index for: ; The historical anomaly correlation index reflects the correlation between the target area and the occurrence of historical anomalies. The larger the value, the stronger the consistency between the target area and the historical anomaly distribution, and the higher the probability of historical recurrence.

[0016] If there are reports of illegal sand mining activities in the target area, and these leads highly overlap with the locations of abnormal areas, it may indicate that the area is related to illegal sand mining activities. This can be determined by calculating the number of reported leads. Calculate the support score for reported clues for: ; in, This indicates the maximum number of reported leads within the region. The higher the value, the stronger the reported leads in that area, and the higher the probability of an anomaly occurring. It should be noted that when lead information is not used, this item can be removed from subsequent risk scoring, and the remaining weights can be re-normalized.

[0017] According to the above technical solution, if the risk score of the suspected abnormal candidate region... Exceeding the preset inspection trigger threshold If the drone is triggered, a rapid drone inspection and verification task will be performed to inspect suspected abnormal candidate areas that meet the preset trigger conditions, in order to identify on-site targets related to illegal sand mining and verify suspected abnormal areas.

[0018] The drone inspection and verification steps include: The drones conducted inspections in areas suspected of being anomalies, acquiring visible light and thermal infrared images. The appearance, texture, and morphological information of visible light images are jointly modeled with the temperature anomaly response information of thermal infrared images using an interactive attention fusion method to obtain a fused feature map. ; By extracting the fusion feature map The confidence level of the identification of sand mining-related targets, thermal anomaly indicators, operation trace scores, and spatial consistency scores are weighted and integrated with the comprehensive risk score to obtain the comprehensive score for UAV verification.

[0019] The work trace score is obtained through the following steps: The lidar point cloud data of the high-risk area to be verified is filtered and denoised, ground points are classified and rasterized; the elevation, slope, surface roughness, normal vector variation, and laser reflection intensity of each raster unit are calculated; based on feature identification, at least one of the following is identified: transport channels, vehicle-trampled areas, ruts, surface disturbance zones, and temporary loading / unloading areas, to obtain the predicted probability and spatial distribution area of ​​each type of work trace; connected component analysis is performed on the identification results, and scattered areas are eliminated based on the area, length, width, aspect ratio, and continuity of the connected components; the work trace identification confidence level is obtained based on the average predicted probability within the effective work trace area. The normalized trace area index is obtained by comparing the total area of ​​the work traces with the area of ​​the reference work traces. The trace continuity index is obtained based on the ratio of the longest continuous centerline length to the total length of all centerlines for linear traces and the ratio of the area of ​​the largest connected region to the total area of ​​the work traces for planar traces. and in accordance with ,in: ; In the formula, Indicates the score for work traces; This indicates the overall identification confidence level of transport routes, vehicle roll zones, ruts, surface disturbance zones, or temporary loading and unloading areas. This represents the total area of ​​work traces identified within the area to be verified; This represents the reference work trace area used for normalizing the work trace area. Its value can be determined based on the typical work trace area in historical verification samples, the statistical results of the work trace area in the candidate area during the current monitoring period, or the preset upper limit of abnormal work area. This represents the normalized trace area index; A trace continuity index, used to characterize the spatial continuity of transport routes, ruts, or surface disturbance zones; These represent the weighting coefficients corresponding to the confidence level of work trace identification, the normalized trace area index, and the trace continuity index, respectively.

[0020] Based on the spatial relationships between the identified on-site work targets such as sand dredging vessels, excavating machinery, transport vehicles, temporary docks, and temporary structures, and the work traces such as transport channels, vehicle traffic zones, and surface disturbance zones, as well as abnormal terrain features such as excavation pits, artificial ramps, work platforms, and temporary stockpiles, the target-trace proximity is calculated. Trace-Topographic Correlation Connectivity with the work chain This leads to the construction of a spatial consistency score: , , , In the formula, This indicates the spatial proximity between the on-site work target and the most recent work trace; N represents the number of identified on-site work targets. This represents the shortest distance between the i-th on-site work target and the nearest work trace; This represents the distance attenuation coefficient, the value of which can be determined based on the scale of typical operating areas, the spatial resolution of point clouds, or the statistical results of historical samples. It indicates the degree of spatial correlation between work traces and abnormal terrain, and is used to characterize whether transport channels, vehicle roll zones or surface disturbance zones are adjacent to, overlap with or connected to excavation pits, artificial ramps, work platforms or temporary stockpiles. Indicates the degree of connectivity in the job chain; This indicates the number of valid relationships that actually satisfy the preset spatial relationships, which include at least one of the following: excavating machinery is adjacent to the excavation pit, transport vehicles are located in the transport channel, the transport channel connects the excavation pit and the temporary stockpile, and sand dredging vessels are adjacent to the temporary dock. This indicates the total number of preset spatial relationships that can be used for judgment within the current area to be verified; These represent the weighting coefficients corresponding to target-trace proximity, trace-terrain correlation, and operation chain connectivity, respectively.

[0021] The sand mining-related targets include sand mining vessels, excavating machinery, transport vehicles, temporary storage yards, temporary docks, transport channels, and traces of abnormal operations.

[0022] The UAV verification process uses visible light and thermal infrared imagery to identify and interpret operational targets and traces within the target area. By comprehensively analyzing the target categories, spatial distribution, and thermal anomaly response characteristics, it determines whether the candidate anomaly areas meet the conditions for further detailed three-dimensional verification and outputs the on-site verification results.

[0023] The specific steps for jointly modeling the appearance, texture, and morphological information of visible light images with the temperature anomaly response information of thermal infrared images using the interactive attention fusion method include: Visible light image features and thermal infrared image features were extracted separately to obtain a visible light feature map. Thermal infrared feature map ; Visible light feature map Thermal infrared feature map Perform linear mappings respectively to obtain the corresponding query matrix, key matrix, and value matrix: , , in, , , , , and This is a trainable mapping matrix. Then, an interactive attention matrix is ​​constructed pointing from visible light to thermal infrared. : ; And construct an interactive attention matrix pointing from thermal infrared to visible light. : ; in, It is a normalization factor used to balance the dimensionality of the feature map. The function is used to normalize the attention weights, where T denotes the transpose. Then, based on the interaction attention matrix, cross-modal enhancement features are calculated respectively: ; ; in, This indicates the enhanced visible light characteristics obtained after being guided by thermal infrared information. This indicates the enhanced thermal infrared features obtained after being guided by visible light information.

[0024] , and These represent the query matrix, key matrix, and value matrix corresponding to visible light features, respectively. , and These represent the query matrix, key matrix, and value matrix corresponding to the thermal infrared features, respectively. The query matrix represents the feature information that the current modality needs to focus on, the key matrix is ​​used for correlation matching with features from another modality, and the value matrix provides the actual feature content participating in cross-modal fusion. Specifically, This represents the visible light value matrix obtained by linear mapping of the visible light feature map, used to provide fusion information such as target appearance, texture, shape, boundary and spatial structure; This represents the thermal infrared value matrix obtained by linear mapping of the thermal infrared feature map, used to provide fusion information such as target temperature anomaly response, heat distribution, and heat intensity. To preserve the original modal features and enhance cross-modal complementarity, a residual fusion method is used to construct the final fused feature map. : ; in, Indicates channel-dimensional splicing. This represents a linear or convolutional mapping function used to remap concatenated features to a predetermined channel dimension. The resulting... It also includes information on the appearance, texture, and shape of visible light images, as well as information on temperature anomaly responses from thermal infrared images.

[0025] According to the above technical solution, the thermal anomaly index : ; ; In the formula, This represents the average radiation temperature of the target area. This represents the median temperature of the neighborhood background. This represents the median absolute deviation of the background temperature. This is the thermal anomaly saturation threshold. This represents the standardized thermal anomaly intensity of a region relative to its surrounding background region, used to measure the degree to which the temperature of the target region is higher than the surrounding background temperature. Where, when When, it indicates that there is no positive thermal anomaly above the background in the target area; when This indicates that there is a certain thermal anomaly in the target area and the thermal anomaly index varies with... Increase and increase; when When the value is saturated, it indicates that there is a significant thermal anomaly in the target area.

[0026] According to the above technical solution, lidar point cloud data of the high-risk area to be verified in three dimensions is obtained, and a refined verification comprehensive index is constructed based on the abnormal terrain features and on-site operation features extracted from the lidar point cloud data. According to the comprehensive index of detailed verification Three-dimensional verification is performed on the high-risk areas to be verified; the comprehensive index for detailed verification is: , In the formula, This represents the comprehensive index of abnormal terrain. This represents the overall index of on-site operations. This indicates the weight corresponding to the comprehensive index of abnormal terrain. This indicates the weight corresponding to the comprehensive index of on-site operations.

[0027] Abnormal terrain features refer to three-dimensional geomorphic features that change abnormally relative to the reference terrain due to mining, stockpiling, mechanical compaction, or temporary construction, including at least one of the following: excavation pits, stockpiles, artificial ramps, work platforms, and bank slope cutting features.

[0028] On-site operation characteristics refer to the target characteristics and operation trace characteristics of sand mining operations extracted from lidar point clouds. The target characteristics of sand mining operations include at least one of sand mining vessels, excavating machinery, transport vehicles, temporary docks, and temporary structures.

[0029] Operation trace features include at least one of the following: transport channels, vehicle crush marks, surface disturbance zones, and spatial correlation features between the target and abnormal terrain.

[0030] According to the above technical solution, the comprehensive index of abnormal terrain is: ; In the formula, The weights corresponding to the volume index of excavated pits within the study area. This indicates the volume of the excavation pit within the study area. This indicates the weight corresponding to the response index. Indicates the response index. This indicates the weight corresponding to the local flatness index. This indicates the local flatness index. The reference maximum excavation volume used to normalize the volume of the excavation pit is determined based on the maximum excavation volume in the historical verification samples of the target river section, the maximum excavation volume of all candidate areas in the current monitoring period, or the preset upper limit of abnormal excavation volume.

[0031] Specifically, a digital terrain model and a digital surface model are generated using the lidar point cloud data; the digital terrain model represents ground elevation, and the digital surface model represents surface elevation. A normalized elevation model is obtained by calculating the difference between ground elevation and surface elevation, which is used to reveal abnormal terrain features. The volume of the excavation pit is then extracted based on the normalized elevation model. : ; In the formula, Indicates the identified excavation pit area. The area of ​​the grid cell. Represents the grid cells within the excavation pit area The elevation difference at a given location is used to characterize the depth of the grid cell relative to the reference terrain.

[0032] Response Index : ; The local flatness index : ; in, Indicates candidate region Internal elevation standard deviation This represents the reference maximum elevation standard deviation used for normalizing the elevation standard deviation. Its value can be determined based on the maximum elevation standard deviation of all candidate areas in the current monitoring period, the maximum elevation standard deviation obtained from historical sample statistics, or a preset upper limit of elevation fluctuation. Represents the raster cells within the candidate region R. The slope response value or terrain gradient response value at a given location is used to characterize the intensity of elevation change near the grid cell; the slope response value can be calculated based on the first-order elevation gradient, slope angle, or neighborhood elevation difference of the current terrain elevation model. This represents the candidate region to be verified in 3D or a local calculation window within the candidate region, determined by suspected abnormal candidate regions, UAV verification target regions, or abnormal terrain patches identified in 3D point clouds. This represents the number of valid raster cells participating in the calculation within the candidate region R.

[0033] According to the above technical solution, the comprehensive index of on-site operations is: , In the formula, This indicates the confidence level in identifying targets during on-site operations. Indicates the work trace index. Indicates the spatial consistency score. This represents the weighting coefficient corresponding to the confidence level of target identification in on-site operations. This represents the weighting coefficient corresponding to the work trace index. This represents the weighting coefficient corresponding to the spatial consistency score.

[0034] Specifically: the lidar point cloud is filtered and denoised, and ground points are classified to generate a current digital terrain model; a reference terrain model is obtained by fitting historical digital terrain models or undisturbed ground points around candidate areas; the elevation difference between the current digital terrain model and the reference terrain model is calculated, and excavation pits and accumulations are identified based on the positive and negative directions, variation range, and spatial connectivity of the elevation difference, and their area, depth, height, and volume are calculated; artificial ramps and work platforms are identified based on the slope, terrain gradient, elevation standard deviation, and continuous area of ​​the current digital terrain model.

[0035] Target point cloud clusters are extracted from the non-ground points after ground point classification. The size, point density, normal vector, linearity, flatness, scattering, and laser reflection intensity of each target point cloud cluster are calculated. Point cloud target recognition models or geometric rule classifiers are used to identify sand dredging vessels, excavating machinery, transport vehicles, temporary docks, and temporary structures, and the target recognition confidence is output. Furthermore, based on the strip morphology, surface roughness, slope, elevation undulation, and spatial connectivity of the ground point cloud, transport channels, vehicle rolling traces, and surface disturbance zones are extracted. Combined with the distance and connection relationship between the on-site operation target and the excavation pit, accumulation body, ramp, and operation platform, a comprehensive index of on-site operation characteristics is obtained.

[0036] This includes a technical solution: a river sand mining anomaly risk assessment system based on multi-source monitoring data, comprising: The multi-source data acquisition and preprocessing module is used to acquire multi-temporal remote sensing data and auxiliary constraint data within the target area and perform preprocessing. The anomaly detection and risk classification module extracts river anomaly features based on preprocessed multi-temporal remote sensing data and auxiliary constraint data, normalizes each river anomaly feature to calculate a comprehensive anomaly index, screens suspected anomaly candidate areas based on the comprehensive anomaly index, and conducts risk assessment on the suspected anomaly candidate areas. The drone inspection and on-site verification module uses drones to inspect and verify suspected abnormal candidate areas, and calculates the comprehensive score of drone verification. The three-dimensional verification and anomaly quantification analysis module determines whether the comprehensive risk score of the suspected anomaly candidate region reaches a high-risk level. And the overall score of drone review When the candidate region is identified as a high-risk region to be verified in three dimensions, three-dimensional verification is performed on the high-risk region to be verified in three dimensions. The inspection record management module is used to structure and update the inspection results corresponding to each candidate abnormal area, forming an abnormal area inspection record.

[0037] Another technical solution includes a storage medium storing at least one instruction, which is loaded and executed by a processor to implement a river sand mining anomaly risk assessment method based on multi-source monitoring data, as described in any of the above technical solutions.

[0038] Compared with the prior art, the beneficial effects achieved by the present invention are: 1. This invention, by constructing a comprehensive anomaly index and introducing a risk triggering mechanism, can improve the timeliness of detecting suspected anomalies and the targeted nature of inspection task organization. Simultaneously, by combining visible light imagery, thermal infrared imagery, and 3D point cloud data, this invention jointly identifies and quantifies on-site operational targets and visible abnormal terrain, enhancing the accuracy and interpretability of auxiliary judgments regarding suspected illegal sand mining activities. This helps reduce misjudgments caused by natural erosion and sedimentation changes, seasonal water level fluctuations, and other non-sand mining activities.

[0039] 2. This invention improves the standardization of result expression and the convenience of subsequent review, continuous tracking and dynamic supervision by outputting the inspection results in a structured manner and forming a continuously updated inspection file of abnormal areas, thereby providing more efficient and reliable technical support for river sand mining supervision.

[0040] 3. This invention extracts river anomaly features based on multi-temporal satellite remote sensing data, UAV remote sensing data, and spatially constrained data. It normalizes each anomaly feature to calculate a comprehensive anomaly index, filters suspected anomaly candidate areas based on the comprehensive anomaly index, and conducts risk assessments on these candidate areas. After UAVs inspect and verify the suspected anomaly candidate areas, high-risk areas confirmed by the UAVs undergo 3D verification. Finally, the results are output and the archive is updated. This invention realizes the screening of suspected anomaly areas in rivers, risk classification, UAV verification, 3D evidence chain verification, and dynamic updating of inspection archives. It can improve the timeliness of discovering suspected illegal sand mining anomalies in rivers, the targeting of inspection tasks, the connection between on-site verification and detailed verification, and the standardization and usability of result output, thus providing more efficient and reliable technical support for river sand mining supervision. Attached Figure Description

[0041] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the steps of a method for assessing the risk of abnormal sand mining in rivers based on multi-source monitoring data according to the present invention; Figure 2 This is a schematic diagram of the structure of a river sand mining anomaly risk assessment system based on multi-source monitoring data according to the present invention. Detailed Implementation

[0042] 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.

[0043] This invention provides a method for assessing the risk of abnormal sand mining in rivers based on multi-source monitoring data. The specific steps include: S1. Acquire multi-temporal remote sensing data and auxiliary constraint data for the target river section at a fixed monitoring cycle. The remote sensing data includes... Synthetic Aperture Radar Imagery and Multispectral optical imaging.

[0044] The auxiliary constraint data includes spatial constraint data and hydrological data. The spatial constraint data includes at least one of the following: river management scope, boundary of prohibited mining area, boundary of key sensitive bank section, historical anomaly distribution information, shoreline vector data, road accessibility information, and permitted scope of legal sand mining or dredging projects. The hydrological data includes at least one of the following: water level, flow rate, rainfall, historical water level process or historical scouring and deposition change information.

[0045] To ensure consistency of multi-source data in terms of spatial location and temporal scale, the optical satellite imagery is first subjected to geometric correction, radiometric correction, cloud and shadow detection, temporal registration, and coordinate unification processing; the synthetic aperture radar imagery is subjected to radiometric calibration, speckle noise suppression, topographic correction, temporal registration, and coordinate unification processing; the hydrological data is subjected to time matching and outlier removal; and the spatially constrained data is standardized, coordinate unified, and spatialized, thereby constructing a monitoring dataset under a unified spatiotemporal reference.

[0046] S2. Based on preprocessed multi-temporal remote sensing data and auxiliary constraint data, extract river channel anomaly features, normalize each river channel anomaly feature, calculate a comprehensive anomaly index, screen suspected anomaly candidate areas based on the comprehensive anomaly index, and conduct risk assessments on the suspected anomaly candidate areas. Specific steps include: S201 and SAR (radar) imagery are highly sensitive to surface roughness, dielectric constant, and water distribution, providing crucial information that is difficult to obtain reliably through optical remote sensing. Suspected illegal sand mining activities and their accompanying mechanical operations may lead to changes in riparian surface roughness, exposed riverbeds, abrupt changes in water extent, and the appearance of temporary structures; therefore, all of these changes can be monitored using SAR features.

[0047] SAR Water Change Index In SAR imagery, water bodies typically exhibit low backscattering characteristics, thus their extent can be extracted using thresholding. SAR water body information is less affected by cloud cover, making it an important supplement to optical water body information, particularly suitable for cloudy and rainy areas. Indicates phase The SAR water change index is calculated based on the water area or the area corresponding to the water boundary extracted from SAR images. ; in, Indicates the previous monitoring phase. Indicates the next monitoring phase. The larger the value of the water change index, the more significant the change in the water body extent between two time phases, and it can be used to characterize the increased degree of disturbance in the target area.

[0048] Texture variation index Sand mining activities and their mechanical operations can also alter the spatial texture structure of the Earth's surface. Texture contrast features can be extracted from SAR imagery using the Gray-Level Co-occurrence Matrix (GLCM), and their temporal changes can be monitored. Let... Indicates phase The texture change index is defined as the mean texture contrast of the target region. ; The larger the value, the more significant the change in surface texture structure, which can be used to characterize local surface structure changes caused by sand pits, deposits, or temporary work areas.

[0049] In monitoring the target river section, in addition to changes in water volume and texture, suspected illegal sand mining activities may also cause shoreline retreat, bank slope disturbance, increased exposure of riverbanks, and damage to riparian vegetation. Therefore, based on the aforementioned anomaly characteristics, shoreline disturbance index, vegetation change index, and turbidity change index are further constructed to improve the ability to identify suspected anomaly areas.

[0050] Shoreline Disturbance Index Suppose a certain target river segment is in the time phase and The widths of the shorelines below are respectively and The shoreline disturbance index Defined as: ; in, Indicates the width of the river channel at the initial time phase. This indicates the width of the river channel in the next time phase. If If the value exceeds a preset threshold, it is determined that there is a strong abnormal change in the shoreline in the area. The threshold can be determined based on the statistical results of historical samples, the optimization results of the validation set, or the results of on-site calibration.

[0051] Vegetation change index Suspected illegal sand mining activities may also damage riparian vegetation systems, causing riparian vegetation damage or a decrease in cover. Therefore, monitoring changes in vegetation cover can be used to assist in the identification of potential sand mining activities. The Normalized Difference Vegetation Index (NDVI) is a commonly used indicator for remote sensing monitoring of vegetation cover, and its calculation formula is as follows: , in and Representing time phase Reflectance in the lower red band and near-infrared band. To characterize the degree of vegetation cover decline, a vegetation change index was further constructed. : , The larger the value, the more significant the decline in vegetation cover in the target area. This phenomenon may be related to sand mining activities and related human activities such as road construction and sediment dumping. It can be used to characterize the degree of vegetation damage caused by road construction, sediment dumping or surface disturbance.

[0052] Turbidity Change Index In certain river sections, the operation of sand dredging machinery or vessels may disturb riverbed sediments, leading to an increase in the concentration of suspended sediment in the local water body, thus causing significant changes in water turbidity within a short period of time. Based on the reflectance of optical image bands, the Normalized Turbidity Index (NDTI) can be calculated. , In the formula, and Representing time phase Reflectance in the red band and reflectance in the green band.

[0053] In monitoring illegal sand mining, the operation of sand mining machinery or vessels disturbs riverbed sediments, causing a short-term increase in turbidity in localized water bodies. Therefore, abnormal disturbance areas can be identified by analyzing the changes in NDTI between two consecutive time phases, and a turbidity change index can be defined. : , The larger the value, the more significant the increase in turbidity of the target area's water body, which can be used to characterize local water body anomalies caused by sand mining operations or mechanical disturbances.

[0054] S202. Based on changes in water body composition, texture, shoreline disturbance, vegetation, and turbidity, and by uniformly normalizing and adaptively weighting and fusing each individual anomaly index, a comprehensive anomaly index is constructed. The comprehensive anomaly index is used to characterize the overall degree of anomaly in the target area during the current monitoring period. The larger the value, the more likely the target area is to have anomalies that require further verification.

[0055] Among them, the individual anomaly indices involved in the fusion include the SAR water change index. Texture variation index Shoreline Disturbance Index Vegetation change index and turbidity change index To ensure the reproducibility of weight calculation, an adaptive weight allocation method based on softmax is adopted. For the ... Individual abnormal index Its gating score is defined as: ; in, This is the proportionality coefficient. It is a bias term, and . No. The weighting coefficients corresponding to each abnormal index Defined as: ; Based on this, the comprehensive anomaly index Defined as: .

[0056] In the formula, , , , , These represent the weights corresponding to the anomaly features of the respective river channels.

[0057] In practical applications, the scaling factor and bias term can be obtained through training with historical labeled samples, or through parameter tuning on the validation set or empirical settings. When sufficient training samples are lacking, a fixed-weight approach can be used as a simplified implementation. This method enhances the system's responsiveness to abnormal changes and improves the accuracy of abnormal region identification.

[0058] S203, Based on the comprehensive anomaly index Anomaly screening is performed on the target river section. Let the anomaly region be defined as the first [section / area] within the target area. The comprehensive anomaly index of the candidate regions is Then when it satisfies At that time, it was identified as a suspected anomaly candidate region. Among them, The anomaly detection threshold can be determined by maximizing the F1 score, Youden index, or combined recall and precision on the validation set based on historical labeled samples. For scenarios where river section monitoring prioritizes controlling missed detections, a lower threshold can also be used. This will improve the recall rate of candidate regions and allow for further screening in subsequent risk grading and drone verification stages.

[0059] S204. Calculate the sensitivity proximity index, historical anomaly correlation index, and reporting clue support score for suspected abnormal candidate regions.

[0060] Wherein, the shortest distance from the candidate region to the boundary of the sensitive region is denoted as . Then define the sensitivity proximity index. : ; in, The distance attenuation coefficient can be determined based on the target river section scale, the width of the management buffer zone, or the statistical results of historical anomaly spatial distribution. The closer the value is to 1, the closer the candidate anomaly region is to the sensitive region, and the higher its potential risk.

[0061] If the target area has historical anomaly records, and anomalies related to sand mining activities have occurred in that area, then the probability of anomalies occurring again in that area is relatively high. Let the frequency of historical anomalies in the spatial unit containing the candidate anomaly area be . The highest frequency of historical anomalies in the study area was Then define the historical anomaly correlation index. for: ; This indicator reflects the correlation between the target area and the occurrence of historical anomalies. The larger the value, the stronger the consistency between the target area and the historical anomaly distribution, and the higher the probability of historical recurrence.

[0062] If there are reports of illegal sand mining activities in the target area, and these leads highly overlap with the locations of abnormal areas, it may indicate that the area is related to illegal sand mining activities. This can be determined by calculating the number of reported leads. Calculate the support score for reported clues for: ; in, This indicates the maximum number of reported leads within the region. The higher the value, the stronger the reported leads in that area, and the higher the probability of an anomaly occurring. It should be noted that when lead information is not used, this item can be removed from subsequent risk scoring, and the remaining weights can be re-normalized.

[0063] S205. A comprehensive risk score is obtained by weighting the comprehensive anomaly index, sensitivity proximity index, historical anomaly correlation index, and the score of reported clues for suspected abnormal candidate areas. : ; in, The weighting coefficients for each anomaly index are adjusted based on expert experience or historical data, and the specific values ​​can be optimized according to the actual situation.

[0064] S206. Based on the comprehensive risk score Further risk classification is then applied to suspected abnormal areas. The formula for risk classification is as follows: ; According to the classification results Areas indicating low risk can be included in routine monitoring or follow-up investigations; For areas designated as medium-risk, drones can be deployed for verification based on available inspection resources. High-risk areas should be prioritized for rapid drone verification missions. The threshold... and It can be determined based on the statistical distribution of historical samples, the results of validation set optimization, or on-site inspection experience.

[0065] S3. Utilize drones to inspect and verify suspected anomaly candidate areas. Specific steps include: S301. For each suspected abnormal area, if the risk score... Exceeding the preset inspection trigger threshold The system will deploy drones for inspection missions. The drones will perform high-resolution imaging of the target area, acquiring visible light and thermal infrared images.

[0066] S302. Jointly model the appearance, texture, and morphological information of visible light images with the temperature anomaly response information of thermal infrared images using an interactive attention fusion method. Specific steps include: Visible light feature map Thermal infrared feature map Perform linear mappings separately to obtain the query matrix, key matrix, and value matrix: ; ; in, , , , , and This is a trainable mapping matrix. Then, an interactive attention matrix is ​​constructed pointing from visible light to thermal infrared. : ; And construct an interactive attention matrix pointing from thermal infrared to visible light. : ; in, It is a normalization factor used to balance the dimensionality of the feature map. The function is used to normalize the attention weights. Then, based on the interaction attention matrix, cross-modal enhancement features are calculated respectively: ; ; in, This indicates the enhanced visible light characteristics obtained after being guided by thermal infrared information. This represents the enhanced thermal infrared features obtained after being guided by visible light information. To preserve the original modal features and enhance cross-modal complementarity, this embodiment employs residual fusion to construct the final fused feature map. : ; in, Indicates channel-dimensional splicing. This represents a linear or convolutional mapping function used to remap concatenated features to a predetermined channel dimension. The resulting... It also includes information on the appearance, texture, and shape of visible light images, as well as information on temperature anomaly responses from thermal infrared images.

[0067] S303, Extracting fused feature maps The identification confidence level, thermal anomaly index, operation trace score, and spatial consistency score of sand mining-related targets were combined with the aforementioned comprehensive risk score to obtain the comprehensive drone verification score for the k-th suspected anomaly candidate area. The identification of sand mining-related targets involves identifying at least one of the following within the target area: sand mining vessels, excavating machinery, transport vehicles, temporary storage yards, temporary docks, transport channels, and abnormal operation traces. in, This indicates the confidence level in identifying sand mining-related targets. Indicates thermal anomaly index, This indicates the score for the work traces. Indicates the spatial consistency score. This represents the comprehensive risk score of the k-th suspected anomaly candidate region. Let each represent the weight of the respective indicator, and satisfy the condition... and .

[0068] To give the thermal anomaly index a clear physical meaning, let the average radiation temperature of the target area be denoted as . The median temperature of its neighborhood background is The median absolute deviation of the background temperature is Then the thermal anomaly index Defined as: ; ; in, This represents the thermal anomaly saturation threshold. When the temperature of the target area is not higher than the background temperature, the thermal anomaly index is set to 0; when the temperature of the target area is higher than the background temperature but has not reached the saturation threshold, the thermal anomaly index increases linearly with the temperature difference; when the intensity of the temperature anomaly reaches or exceeds the saturation threshold, the thermal anomaly index is set to 1.

[0069] The work trace score is obtained through the following steps: The lidar point cloud data of the high-risk area to be verified is filtered and denoised, ground points are classified and rasterized; the elevation, slope, surface roughness, normal vector variation, and laser reflection intensity of each raster unit are calculated; based on feature identification, at least one of the following is identified: transport channels, vehicle-trampled areas, ruts, surface disturbance zones, and temporary loading / unloading areas, to obtain the predicted probability and spatial distribution area of ​​each type of work trace; connected component analysis is performed on the identification results, and scattered areas are eliminated based on the area, length, width, aspect ratio, and continuity of the connected components; the work trace identification confidence level is obtained based on the average predicted probability within the effective work trace area. The normalized trace area index is obtained by comparing the total area of ​​the work traces with the area of ​​the reference work traces. The trace continuity index is obtained based on the ratio of the longest continuous centerline length to the total length of all centerlines for linear traces and the ratio of the area of ​​the largest connected region to the total area of ​​the work traces for planar traces. and in accordance with ,in: , In the formula, Indicates the score for work traces; This indicates the overall identification confidence level of transport routes, vehicle roll zones, ruts, surface disturbance zones, or temporary loading and unloading areas. This represents the total area of ​​work traces identified within the area to be verified; This represents the reference work trace area used for normalizing the work trace area. Its value can be determined based on the typical work trace area in historical verification samples, the statistical results of the work trace area in the candidate area during the current monitoring period, or the preset upper limit of abnormal work area. This represents the normalized trace area index; A trace continuity index, used to characterize the spatial continuity of transport routes, ruts, or surface disturbance zones; These represent the weighting coefficients corresponding to the confidence level of work trace identification, the normalized trace area index, and the trace continuity index, respectively.

[0070] Based on the spatial relationships between the identified on-site work targets such as sand dredging vessels, excavating machinery, transport vehicles, temporary docks, and temporary structures, and the work traces such as transport channels, vehicle traffic zones, and surface disturbance zones, as well as abnormal terrain features such as excavation pits, artificial ramps, work platforms, and temporary stockpiles, the target-trace proximity is calculated. Trace-Topographic Correlation Connectivity with the work chain This leads to the construction of a spatial consistency score: , , , In the formula This indicates the spatial proximity between the on-site work target and the most recent work trace; N represents the number of identified on-site work targets. This represents the shortest distance between the i-th on-site work target and the nearest work trace; This represents the distance attenuation coefficient, the value of which can be determined based on the scale of typical operating areas, the spatial resolution of point clouds, or the statistical results of historical samples. It indicates the degree of spatial correlation between work traces and abnormal terrain, and is used to characterize whether transport channels, vehicle roll zones or surface disturbance zones are adjacent to, overlap with or connected to excavation pits, artificial ramps, work platforms or temporary stockpiles. Indicates the degree of connectivity in the job chain; This indicates the number of valid relationships that actually satisfy the preset spatial relationships, which include at least one of the following: excavating machinery is adjacent to the excavation pit, transport vehicles are located in the transport channel, the transport channel connects the excavation pit and the temporary stockpile, and sand dredging vessels are adjacent to the temporary dock. This indicates the total number of preset spatial relationships that can be used for judgment within the current area to be verified; These represent the weighting coefficients corresponding to target-trace proximity, trace-terrain correlation, and operation chain connectivity, respectively.

[0071] Combining the target's spatial location, morphological characteristics, historical observation data, and temperature anomaly information from thermal infrared images, an indicator for determining illegal sand mining is formed based on the results of on-site verification by drones. This is used to determine whether the candidate anomaly region meets the conditions for entering the three-dimensional fine verification stage.

[0072] S4. Set the trigger threshold for 3D fine-tuning verification. When the comprehensive risk score of suspected abnormal candidate regions... Furthermore, the overall score for drone review meets the requirements. The candidate area was identified as a high-risk area requiring 3D verification, and a detailed 3D verification task using UAV lidar or photogrammetry was initiated. In the absence of historical 3D verification samples for the target river section, the expert experience of river sand mining supervisors, UAV inspectors, and remote sensing interpreters could be used to further refine the verification process. The initial value is set to 0.65. This value indicates that the UAV verification results have reached a medium-to-high level of anomaly support. After accumulating local measured data, the threshold is calibrated using historical samples that simultaneously contain UAV verification results, 3D verification results, and on-site investigation results. Specifically, based on the 3D verification or on-site investigation results, the historical samples are divided into anomaly confirmation samples and anomaly exclusion samples. By comparing the anomaly recall rate, false negative rate, and false trigger rate under different candidate thresholds, the updated threshold is determined. In regulatory scenarios where the priority is to avoid missed detections, the threshold with the lowest false trigger rate when the recall rate of confirmed abnormal samples reaches the preset requirements can be selected.

[0073] Acquire lidar point cloud data of the high-risk area to be verified in three dimensions, and construct a refined verification comprehensive index based on the abnormal terrain features and on-site operation features extracted from the lidar point cloud data. According to the comprehensive index of detailed verification Three-dimensional verification is performed on the high-risk areas to be verified; the comprehensive index for detailed verification is: , In the formula, This represents the comprehensive index of abnormal terrain. This represents the overall index of on-site operations. This indicates the weight corresponding to the comprehensive index of abnormal terrain. This indicates the weight corresponding to the comprehensive index of on-site operations.

[0074] The comprehensive index of abnormal terrain is: ; In the formula, The weights corresponding to the volume index of excavated pits within the study area. This indicates the volume of the excavation pit within the study area. This indicates the weight corresponding to the response index. Indicates the response index. This indicates the weight corresponding to the local flatness index. This indicates the local flatness index.

[0075] This process involves acquiring lidar point cloud data of high-risk areas awaiting review, extracting anomalous terrain features such as excavation pits, stockpiles, ramps, and work platforms through 3D verification and anomaly quantification analysis, and quantifying the area, depth, height, and volume of these anomalous areas to provide quantifiable evidence for determining illegal sand mining activities. Specifically: First, acquire lidar point cloud data of the high-risk area to be reviewed, perform point cloud filtering and noise reduction, and then extract the ground point set using a cloth-simulation filtering algorithm. Non-ground point set .

[0076] A digital terrain model (DEM) and a digital surface model (DSM) are generated using lidar point cloud data to represent ground elevation and surface elevation, respectively. A normalized elevation model is obtained by calculating the difference between the two models to reveal anomalous terrain features and subsequently extract the volume of the excavation pit. : ; in, Indicates the identified excavation pit area. This represents the area of ​​a grid cell.

[0077] When the normalized elevation model shows significant changes, the area can be identified as an excavation pit or a stockpile. Specifically, when the elevation change at a certain location exceeds a set threshold, the area is marked as an excavation pit or a stockpile, and the area, depth, and volume of the abnormal area are then calculated. The volume of an excavation pit can be estimated by multiplying the elevation change by the area of ​​the raster cell; the volume of a stockpile is calculated in a similar way.

[0078] Besides excavation pits and stockpiles, illegal sand mining activities often create artificial terrain features such as ramps and work platforms. The slope values ​​corresponding to changes in elevation are calculated. This allows for further identification of these terrain features. The formula for calculating slope is: ; ; To facilitate subsequent weighted integration with indicators such as excavation pit volume and local flatness, the slope response is currently normalized. Indicates the magnitude of the original terrain gradient. This indicates the maximum slope value in the area. This represents the normalized slope response, with values ​​ranging from 0 to 1. When using a projected coordinate system and defining the positive y-axis as east and the positive y-axis as north,... Describe the elevation changes in the east-west direction. Describe the elevation changes in the north-south direction.

[0079] When the slope of a certain area changes continuously and it has a strong spatial connection with the excavation pit and the stockpile, this area can be identified as a potential transport ramp. Therefore, candidate areas are defined. Response Index for: ; Furthermore, if an area has relatively flat elevation with minimal local elevation changes, and is adjacent to a storage area or access path, it can be identified as a suspected work platform. Local flatness index Defined as: ; in, Indicates candidate region Internal elevation standard deviation. The greater the local flatness, the flatter the area, and the more suitable it is to be set up as a work platform.

[0080] The comprehensive index of on-site operations is: , In the formula, This indicates the confidence level in identifying targets during on-site operations. Indicates the work trace index. Indicates the spatial consistency score. This represents the weighting coefficient corresponding to the confidence level of target identification in on-site operations. This represents the weighting coefficient corresponding to the work trace index. This represents the weighting coefficient corresponding to the spatial consistency score.

[0081] Based on this, the lidar point cloud is filtered and denoised, and ground points are classified. The current digital terrain model is generated using the ground points. A reference terrain model is obtained by fitting historical digital terrain models or undisturbed ground points around candidate areas. The elevation difference between the current digital terrain model and the reference terrain model is calculated. Based on the positive and negative directions, variation range, and spatial connectivity of the elevation difference, excavation pits and accumulation bodies are identified, and their area, depth, height, and volume are calculated. Based on the slope, terrain gradient, elevation standard deviation, and continuous area of ​​the current digital terrain model, artificial ramps and work platforms are identified.

[0082] Target point cloud clusters are extracted from the non-ground points after ground point classification. The size, point density, normal vector, linearity, flatness, scattering, and laser reflection intensity of each target point cloud cluster are calculated. Point cloud target recognition models or geometric rule classifiers are used to identify sand dredging vessels, excavating machinery, transport vehicles, temporary docks, and temporary structures, and the target recognition confidence is output. Furthermore, based on the strip morphology, surface roughness, slope, elevation undulation, and spatial connectivity of the ground point cloud, transport channels, vehicle rolling traces, and surface disturbance zones are extracted. Combined with the distance and connection relationship between the on-site operation target and the excavation pit, accumulation body, ramp, and operation platform, a comprehensive index of on-site operation characteristics is obtained.

[0083] S5. Result Output and Inspection File Generation. The system integrates the anomaly index, risk classification results, drone rapid verification results, and 3D detailed verification results to form a structured inspection output for candidate anomaly areas, and updates the anomaly area inspection files for subsequent review, continuous tracking, and regulatory applications.

[0084] For candidate anomaly areas that require continued monitoring after review and verification, corresponding inspection results are output. These results include the spatial location of the anomaly area, risk level, task trigger time, UAV review time, on-site target category, target detection confidence level, thermal anomaly index, three-dimensional quantification results, and auxiliary judgment results. The three-dimensional quantification results include at least one of the following: anomaly excavation area, accumulation area, anomaly excavation volume, accumulation volume, transport ramp connectivity index, local flatness index, and three-dimensional fine verification comprehensive index.

[0085] An inspection file is established for each candidate anomaly area. This file includes the anomaly area's unique identifier, spatial location, scoring results at each stage, key timestamps, on-site target information, quantitative results of the anomaly terrain, verification notes, and status update records. The inspection files for each anomaly area are compiled into an audit record database, and inspection process information, anomaly identification results, UAV verification results, 3D detailed verification results, and historical records are uniformly stored, updated, and managed.

[0086] During subsequent inspections, if candidate anomaly areas are identified again at the same spatial location or within adjacent spatial areas, or if the risk status of existing candidate anomaly areas changes, the corresponding inspection records will be updated. By continuously updating the inspection records, a time-series evolution record of anomaly areas in the target river section is formed, providing a technical basis for risk retrospective analysis, key area control, and regulatory decision-making.

[0087] Based on cloud cover, shadow ratio, imaging time difference, SAR image speckle noise level, temporal registration error, and hydrological data time matching degree, a reliability coefficient is calculated for the corresponding data source or anomaly feature. This reliability coefficient is used for adaptive weighting in the subsequent construction of the comprehensive anomaly index to reduce the impact of low-quality observation data on anomaly screening results.

[0088] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0089] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing the risk of abnormal sand mining in river channels based on multi-source monitoring data, characterized in that, The steps include: Acquire multi-temporal remote sensing data and auxiliary constraint data within the target area and perform preprocessing; River channel anomaly features are extracted based on preprocessed multi-temporal remote sensing data and auxiliary constraint data. The anomaly features of each river channel are normalized to calculate the comprehensive anomaly index. Suspected anomaly candidate areas are screened based on the comprehensive anomaly index and risk assessment is conducted on the suspected anomaly candidate areas. Drones were used to inspect and verify suspected anomaly candidate areas, and the overall score of the drone verification was calculated. When the comprehensive risk score of suspected abnormal candidate areas And the overall score of drone review When the candidate region is identified as a high-risk region to be verified in three dimensions, three-dimensional verification is performed on the high-risk region to be verified in three dimensions. The inspection results corresponding to each candidate abnormal area are stored and updated in a structured manner to form an abnormal area inspection file.

2. The method for assessing the risk of abnormal river sand mining based on multi-source monitoring data according to claim 1, characterized in that, An adaptive weight allocation method based on softmax is used to determine the weight of each of the river channel anomaly features. The river channel anomaly features are then normalized and weighted and fused to obtain a comprehensive anomaly index. The abnormal features of the river channel include water boundary change index, shoreline disturbance index, exposed surface change index, vegetation change index, water turbidity change index, suspected storage yard change index, transportation channel change index, and temporary wharf change index.

3. The method for assessing the risk of abnormal river sand mining based on multi-source monitoring data according to claim 2, characterized in that, When the target area is the The comprehensive anomaly index of each candidate region At that time, it was identified as a suspected anomaly candidate region; among them, The threshold for anomaly identification is set; a risk assessment is performed on the suspected anomaly candidate regions, the steps of which include: Calculate the sensitivity proximity index, historical anomaly correlation index, and whistleblower support score for suspected anomaly candidate regions; A comprehensive risk score is obtained by weighting the comprehensive anomaly index, sensitivity proximity index, historical anomaly correlation index, and the score supporting reported clues for suspected anomaly candidate regions. ; Based on comprehensive risk score Risk classification for suspected abnormal areas: ; In the formula, Indicates a low-risk area. This indicates a medium-risk area. Indicates a high-risk area. This indicates the first risk threshold. This indicates the second risk threshold.

4. The method for assessing the risk of abnormal river sand mining based on multi-source monitoring data according to claim 3, characterized in that, If the comprehensive risk score of the suspected abnormal candidate region Exceeding the preset inspection trigger threshold This will trigger the drone inspection and verification task; The drone inspection and verification steps include: The drones conducted inspections in areas suspected of being anomalies, acquiring visible light and thermal infrared images. The appearance, texture, and morphological information of visible light images are jointly modeled with the temperature anomaly response information of thermal infrared images using an interactive attention fusion method to obtain a fused feature map. ; Using object recognition networks to identify fused feature maps The types, locations, and confidence levels of sand mining targets; the types of sand mining targets include sand mining vessels, excavating machinery, transport vehicles, temporary storage yards, temporary docks, transport channels, and abnormal operation traces; By extracting the fusion feature map The confidence level of the identification of sand mining-related targets, thermal anomaly indicators, operation trace scores, and spatial consistency scores are weighted and integrated with the comprehensive risk score to obtain the comprehensive score for UAV verification.

5. The method for assessing the risk of abnormal river sand mining based on multi-source monitoring data according to claim 4, characterized in that, The thermal anomaly index : ； ； In the formula, This represents the average radiation temperature of the target area. This represents the median temperature of the neighborhood background. This represents the median absolute deviation of the background temperature. This is the thermal anomaly saturation threshold. This represents the normalized thermal anomaly intensity of the target region relative to the surrounding background region.

6. The method for assessing the risk of abnormal river sand mining based on multi-source monitoring data according to claim 1, characterized in that, Acquire lidar point cloud data of the high-risk area to be verified in three dimensions, and construct a refined verification comprehensive index based on the abnormal terrain features and on-site operation features extracted from the lidar point cloud data. According to the comprehensive index of detailed verification Three-dimensional verification is performed on the high-risk areas to be verified; the comprehensive index for detailed verification is: ； In the formula, This represents the comprehensive index of abnormal terrain. This represents the overall index of on-site operations. This indicates the weight corresponding to the comprehensive index of abnormal terrain. This indicates the weight corresponding to the comprehensive index of on-site operations.

7. The method for assessing the risk of abnormal river sand mining based on multi-source monitoring data according to claim 6, characterized in that, The comprehensive index of the abnormal terrain is: ； In the formula, The weights corresponding to the volume index of excavated pits within the study area. This indicates the volume of the excavation pit within the study area. This indicates the weight corresponding to the response index. Indicates the response index. This indicates the weight corresponding to the local flatness index. This indicates the local flatness index.

8. The method for assessing the risk of abnormal river sand mining based on multi-source monitoring data according to claim 6, characterized in that: The comprehensive index for on-site operations is: ； In the formula, This indicates the confidence level in identifying targets during on-site operations. Indicates the work trace index. Indicates the spatial consistency score. This represents the weighting coefficient corresponding to the confidence level of target identification in on-site operations. This represents the weighting coefficient corresponding to the work trace index. This represents the weighting coefficient corresponding to the spatial consistency score.

9. A river sand mining anomaly risk assessment system based on multi-source monitoring data, characterized in that, include: The multi-source data acquisition and preprocessing module is used to acquire multi-temporal remote sensing data and auxiliary constraint data within the target area and perform preprocessing. The anomaly detection and risk classification module extracts river anomaly features based on preprocessed multi-temporal remote sensing data and auxiliary constraint data, normalizes each river anomaly feature to calculate a comprehensive anomaly index, screens suspected anomaly candidate areas based on the comprehensive anomaly index, and conducts risk assessment on the suspected anomaly candidate areas. The drone inspection and on-site verification module uses drones to inspect and verify suspected abnormal candidate areas, and calculates the comprehensive score of drone verification. The 3D verification and anomaly quantification analysis module calculates the comprehensive risk score of suspected anomaly candidate areas. And the overall score of drone review When the candidate region is identified as a high-risk region to be verified in three dimensions, three-dimensional verification is performed on the high-risk region to be verified in three dimensions. The inspection record management module is used to structure and update the inspection results corresponding to each candidate abnormal area, forming an abnormal area inspection record.

10. A storage medium, characterized in that, The storage medium stores at least one instruction, which is loaded and executed by a processor to implement the method as described in any one of claims 1-8.

Images