Soil heavy metal portable XRF intelligent detection and early warning system integrated with risk assessment function

Abstract

The invention discloses a soil heavy metal portable XRF intelligent detection and early warning system integrated with a risk assessment function, and relates to the technical field of environment monitoring. According to the invention, data are collected through a portable equipment system integrating multispectral sensing and an environment probe, a pollution area is accurately locked in combination with an improved clustering algorithm, accurate positioning of a pollution point location is realized by means of cooperative operation of the unmanned aerial vehicle, and the problem of blindness of traditional sampling is effectively solved; a dynamic interference correction and multi-stage calibration mechanism is established, a deep learning and machine learning fusion model is adopted to extract an environmental interference factor, and spectral line processing and an environmental compensation algorithm are combined, so that the accuracy of a detection result in a complex environment is remarkably improved; the efficiency and precision of soil heavy metal pollution treatment are greatly improved by realizing full-process intelligence from risk evaluation to treatment decision, establishing a multi-dimensional risk evaluation system, automatically generating an optimal operation path and intelligently recommending a treatment scheme based on pollution characteristic data.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, specifically a portable XRF intelligent detection and early warning system for heavy metals in soil that integrates risk assessment functions. Background Technology

[0002] Soil environmental monitoring and remediation are crucial for ensuring ecological security and human health, directly impacting food security, sustainable land use, and public well-being. With rapid industrialization and urbanization, soil heavy metal pollution has become increasingly severe, a critical issue in environmental science and technology research. Soil pollution not only affects crop growth but can also accumulate through the food chain, threatening human health. Therefore, precise monitoring and timely intervention are particularly urgent. However, current soil monitoring methods have significant shortcomings in practical application. Many traditional methods are often limited to single-stage detection, lacking the ability to integrate the entire process from sampling to risk assessment. They are particularly inadequate in terms of adaptability and real-time performance in complex environments. Existing equipment often relies on human experience to determine sampling locations, easily missing key pollution areas. Furthermore, their ability to correct for environmental interference in data processing is limited, leading to significant biases in results. These problems make it difficult for soil pollution monitoring to meet the needs of rapid response and precise intervention.

[0003] Against this backdrop, the field of soil monitoring faces significant technical challenges. One core difficulty lies in achieving accurate sampling location and sample representativeness in complex field environments. The diversity of soil types and pollution distribution necessitates that equipment possess rapid sensing capabilities for regional pollution characteristics; otherwise, sampling bias may affect the accuracy of subsequent analyses. Furthermore, this insufficient sensing capability leads to another closely related challenge: how to perform real-time correction and analysis of collected data at the equipment end to address interference from environmental factors such as soil moisture and texture. If these interfering factors cannot be quickly processed on-site, the monitoring results will fail to reflect the true pollution situation. For example, in a farmland, equipment may miss key sampling points due to inaccurate identification of highly polluted areas, or the detection values ​​may be distorted due to failure to consider the influence of soil moisture, ultimately affecting the reliability of pollution assessments. Summary of the Invention

[0004] The purpose of this invention is to provide a portable XRF intelligent detection and early warning system for heavy metals in soil that integrates risk assessment functions, thereby realizing intelligent management of the entire process of heavy metal detection and remediation in soil.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] This application provides a portable XRF intelligent detection and early warning system for heavy metals in soil with integrated risk assessment function, including a data acquisition and integration module, a pollution hotspot area identification module, an air-ground collaborative positioning module, a detection signal dynamic correction module, a pollution concentration calibration and calculation module, a sampling path optimization and planning module, and a remediation scheme real-time recommendation module.

[0007] The data acquisition and integration module, through the deployment of a portable XRF front-end module integrating multispectral sensors and miniature environmental probes, collects characteristic spectral data of heavy metals in the soil of the target area and multidimensional environmental parameters, forming an initial environmental feature set in two dimensions: spectral and environmental.

[0008] The pollution hotspot identification module, based on the initial environmental feature set, adopts an improved density clustering algorithm, incorporates regional prior knowledge, groups soil types, and associates the probability of heavy metal pollution to identify high-confidence potential pollution hotspots.

[0009] The air-ground collaborative positioning module, if the concentration in a potential pollution hotspot area exceeds the preset risk threshold, triggers a collaborative mechanism between the portable XRF detection device and the drone to acquire high-resolution soil texture images and rough heavy metal concentration data, and then guides the portable XRF detection device to locate a highly polluted and highly representative point.

[0010] The detection signal dynamic correction module uses a convolutional neural network-random forest fusion model to dynamically correct interference in the XRF detection signal based on soil texture data and previous environmental parameters.

[0011] The pollution concentration calibration calculation module substitutes the dynamic interference correction coefficient into the initial environmental feature set and combines the heavy metal characteristic spectral line baseline correction of portable XRF technology to obtain a calibrated pollution concentration estimate that is close to the true value.

[0012] The sampling route optimization and planning module compares the calibrated pollution concentration estimate with the preset pollution standard, introduces a risk classification model, generates a sampling route optimization sequence driven by two factors: risk level and geographical accessibility, and outputs the optimal sampling and intervention points.

[0013] The real-time governance solution recommendation module, based on sampling and intervention sites, accurately extracts the characteristic spectral lines of target heavy metal ions using portable XRF, combines previous environmental data, calls upon a differentiated governance solution library, and generates a real-time governance recommendation sequence. The recommendation results are also associated with the risk level of the site.

[0014] The beneficial effects of this invention are as follows:

[0015] By constructing a multi-source data acquisition and intelligent identification system that integrates air and ground, the problems of blind sampling and insufficient representativeness in traditional soil monitoring have been solved. Using portable devices that integrate multispectral sensors and environmental probes, the system collects spectral and environmental parameters. Through an improved clustering algorithm that integrates prior knowledge, pollution hotspots are accurately located. When an abnormal area is detected, drones are automatically activated for collaborative operation. Combined with image recognition and concentration scanning technology, the pollution points are accurately located, fundamentally changing the limitations of relying on human experience and ensuring the scientific nature and representativeness of the sampling points.

[0016] By establishing a dynamic interference correction and multi-level calibration mechanism, the interference of complex environmental factors on the detection results is effectively overcome. A model integrating deep learning and machine learning is used to automatically extract key environmental factors from soil texture and quantify the degree of interference of each factor on the detection signal. Combined with spectral line noise reduction and baseline correction technology, feature signals are accurately extracted. Then, through environmental compensation algorithms and historical calibration rules, the detection results are optimized in multiple levels, which significantly improves the detection accuracy in complex field environments and makes the final results closer to the real pollution situation.

[0017] By achieving intelligent management of the entire process from risk assessment to remediation decision-making, the traditional segmented operation mode has been completely transformed. It intelligently compares precise pollution concentration data with national standards to establish a multi-dimensional risk assessment system; integrates geospatial factors to automatically generate optimal sampling and intervention paths; and finally, based on detailed pollution characteristic data, it intelligently matches and recommends personalized remediation solutions from a knowledge base, achieving seamless integration from detection and discovery to remediation decision-making, and significantly improving the overall efficiency and accuracy of soil pollution remediation. Attached Figure Description

[0018] To better understand and implement this application, the technical solution is described in detail below with reference to the accompanying drawings.

[0019] Figure 1 A schematic diagram of a portable XRF intelligent detection and early warning system for heavy metals in soil with integrated risk assessment function provided in Embodiment 1 of this application;

[0020] Figure 2 A flowchart illustrating the air-ground collaborative positioning module in a portable XRF intelligent detection and early warning system for heavy metals in soil with integrated risk assessment function, provided in Embodiment 1 of this application;

[0021] Figure 3 This is a flowchart illustrating the sampling path optimization and planning module in a portable XRF intelligent detection and early warning system for heavy metals in soil with integrated risk assessment function, as provided in Embodiment 1 of this application. Detailed Implementation

[0022] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, exemplary embodiments will be described in detail below, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0023] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0024] The following detailed description of the specific implementation methods, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided in detail.

[0025] Example 1

[0026] Please see Figures 1-3 This embodiment provides a portable XRF intelligent detection and early warning system for heavy metals in soil with integrated risk assessment function, including:

[0027] The data acquisition and integration module, through the deployment of a portable XRF front-end module integrating multispectral sensors and miniature environmental probes, collects characteristic spectral data of heavy metals in the soil of the target area and multidimensional environmental parameters, forming an initial environmental feature set in two dimensions: spectral and environmental.

[0028] Furthermore, the data acquisition and integration module includes:

[0029] The portable XRF front-end module is activated to synchronously excite and collect the heavy metal characteristic spectral signals of the soil at preset grid points within the target area. At the same time, a miniature environmental probe is used to record multi-dimensional environmental parameters, including at least soil moisture, pH value and temperature.

[0030] The preset grid points are laid out according to the technical specifications for soil environmental monitoring and in combination with the topography of the target area. The grid accuracy can be adjusted between 10m×10m and 50m×50m depending on the survey scale. The portable XRF front-end module excites the soil sample with X-rays for 3 seconds at the center of each grid point through tube voltage and automatically adjusted tube current, and simultaneously collects the Kα and Lα characteristic spectral signals of key heavy metal elements such as lead, arsenic, chromium, cadmium and mercury, which are sensed by the silicon drift detector.

[0031] At the same time as X-ray excitation, the miniature environmental probe simultaneously measures and records the volumetric water content, pH value, and temperature parameters at a depth of 10 cm at the location by inserting a probe into the soil, ensuring that the spectral data and environmental parameters are strictly correlated in time and space. All collected data are accompanied by high-precision GPS / RTK positioning coordinates and timestamps, forming a raw data package with spatiotemporal identification.

[0032] The collected raw spectral data and environmental parameter data were standardized in format and dimension, respectively, and noise was filtered by wavelet transform algorithm to obtain clean spectral datasets and environmental datasets.

[0033] The preprocessing includes: converting raw data from different sensors into a common data format; normalizing the intensity values ​​of spectral data to the [0,1] interval; unifying the units of soil moisture as volume percentage, pH as dimensionless, and temperature as degrees Celsius for environmental parameters; using wavelet transform algorithm to filter noise from the normalized spectral data; selecting the sym4 wavelet basis function to perform 5-level decomposition of the spectral signal; processing the high-frequency detail coefficients of each decomposition layer using the Stein unbiased risk estimation threshold; retaining the low-frequency approximation coefficients that represent the true signal; finally, obtaining the denoised clean spectral data through wavelet reconstruction; and then using the moving average method for smoothing, with the window size set to 5 adjacent sampling points to eliminate random fluctuations and ensure the spatial continuity of the data.

[0034] If there are missing values ​​in the preprocessed dataset, Kriging space interpolation is used to complete them, forming a complete spatial dataset. Principal component analysis is then used to reduce the dimensionality of the completed data, extracting key principal components whose contribution rate exceeds a preset threshold, thus forming a simplified feature combination.

[0035] The process employs Kriging spatial interpolation for data completion. Centering on points with missing values, all known data points within a 50-meter radius are searched as interpolation samples. Based on the spatial autocorrelation of these sample points, a semi-variogram model is fitted, preferably an exponential model. Subsequently, the optimal unbiased estimate of the missing points is calculated using the semi-variogram model, ultimately generating a complete, missing-value-free spatial dataset. This complete spatial dataset is organized into a sample-variable matrix and standardized to eliminate the influence of dimensions. The covariance matrix, eigenvalues, and eigenvectors of this matrix are calculated. The eigenvalues ​​are sorted in descending order, and their cumulative contribution rate is calculated. A cumulative contribution rate threshold of at least 85% is set, and the eigenvectors corresponding to the top k eigenvalues ​​are selected as principal component directions. The original data is projected onto the selected k principal component directions to obtain a set of new, uncorrelated feature variables with significantly reduced dimensionality, i.e., key principal components. These key principal components constitute a concise feature combination capable of representing most of the information in the original data, used for the subsequent construction of the environmental feature set.

[0036] Based on the simplified feature combination, a structured initial environmental feature set is constructed. The feature set characterizes the spatial distribution characteristics of heavy metals in the soil within the target area. Finally, it is compressed and encrypted and stored in a preset cloud or local database, and each dataset is assigned a unique identifier for subsequent modules to call.

[0037] Specifically, by deploying a portable XRF front-end module integrating multispectral sensors and miniature environmental probes, heavy metal characteristic spectral signals and multidimensional environmental parameters are synchronously collected at standardized grid points within the target area. Wavelet transform noise reduction, Kriging space interpolation, and principal component analysis are used to construct an initial environmental feature set with both spectral and environmental dimensions. This solves the problems of single data collection and insufficient consideration of environmental interference factors in traditional soil monitoring methods, which lead to large deviations in detection results. It achieves standardized integration of multi-source heterogeneous data and efficient extraction of key features, providing a high-quality data foundation for subsequent accurate risk assessment and pollution location.

[0038] The pollution hotspot identification module, based on the initial environmental feature set, uses an improved density clustering algorithm, incorporates prior knowledge such as regional soil type distribution and historical pollution data, and groups soil types while associating them with the probability of heavy metal pollution to identify high-confidence potential pollution hotspots.

[0039] Furthermore, the pollution hotspot area identification module includes:

[0040] The system calls up the pre-stored regional soil type distribution map and historical pollution records, spatially correlates them with the initial environmental feature set, and constructs a multi-dimensional prior information database. An improved density clustering algorithm is used to optimize the distance metric by introducing soil type constraints, clustering spatial points to achieve the division of soil category groups, and simultaneously calculating the average pollution probability of each cluster.

[0041] The improved density clustering algorithm incorporates an improved distance metric function to account for the impact of soil type differences on the clustering results; specifically, it is expressed as follows: ; It is the Euclidean distance between points i and j; It is an indicator function that takes the value 1 when the soil types of points i and j are different, and 0 otherwise; It is the type penalty factor, an adjustable parameter greater than 0, used to control the degree of influence of soil type differences on the clustering results.

[0042] Furthermore, the improved density clustering algorithm also includes: adopting a dynamic parameter optimization strategy, automatically determining the optimal neighborhood radius ε through the elbow rule, calculating the rate of change of the number of clusters under different ε values, and selecting the minimum ε value corresponding to the rate of change ≤ 5%; adaptively calculating the minimum number of samples MinPts based on the data distribution characteristics, with the formula: MinPts = ln(N) × d, where N is the total number of samples and d is the feature dimension; determining the soil type penalty factor α through grid search, testing the cluster profile coefficient in the interval [0.1, 1.0] with a step size of 0.1, and selecting the α value that maximizes the profile coefficient; introducing an adaptive parameter adjustment mechanism, automatically increasing the α value by 0.1 steps and re-clustering when the cluster purity (the proportion of the same soil type in the cluster) is lower than 85%.

[0043] A preliminary pollution risk distribution map is generated based on the clustering results; outliers are filtered and removed by setting a preset outlier detection threshold (e.g., Z-score>3) to obtain a corrected risk distribution dataset; and potential pollution hotspots are sorted by setting a high confidence standard (e.g., pollution probability is higher than 80%) to generate a priority list.

[0044] The priority list is visualized to generate a regional distribution map. Using the spatial overlay analysis tool in the geographic information system, the boundaries of soil type zones are accurately matched with potential pollution hotspots to identify key areas of concern with high overlap between the two.

[0045] By applying the pollution screening method and based on the principle of heavy metal synergistic effect, overlapping areas are filtered twice to finally identify a set of high-risk pollution areas and output their geographical boundary coordinates.

[0046] The decision rules for risk distribution correction and inventory generation are as follows: After generating a preliminary pollution risk distribution map, the Z-score method is used to identify outliers: the Z-score value of the pollution probability of each point is calculated, and if its absolute value is greater than 3, it is judged as a statistical outlier and removed. For the corrected dataset, a risk confidence standard is set: areas with a pollution probability higher than 80% are marked as high-confidence potential pollution hotspots. Priority ranking is based on the comprehensive score of two indicators: the primary indicator is the pollution probability, and the secondary indicator is the sample size of the cluster to which the point belongs, representing the spatial representativeness of the area.

[0047] The principle for identifying key areas of concern with high overlap is as follows: using the Intersect spatial analysis tool in the geographic information system, the vector boundary of the soil type zone is overlaid with the potential pollution hotspot. The identified key areas of concern must meet the condition that the overlap area ratio exceeds 70%. A secondary filtering method is applied using pollution screening: based on the principle of heavy metal synergistic effect, if a certain area has compound pollution of three or more heavy metals such as lead, arsenic and cadmium, its risk level is automatically upgraded by one level, and it is finally locked as a high-risk pollution area, and its minimum bounding rectangle or convex hull boundary coordinates are output.

[0048] Specifically, by integrating prior knowledge such as soil type distribution and historical pollution data, and employing an improved density clustering algorithm that incorporates soil type constraints, this method not only achieves scientific grouping of soil categories but also accurately correlates the probability of heavy metal pollution. Combined with outlier screening, spatial overlay analysis, and heavy metal synergistic effect assessment, it solves the problems of insufficient accuracy in identifying potential pollution hotspots and inadequate spatial representativeness in complex pollution environments using traditional methods. Ultimately, it achieves accurate identification and priority ranking of high-confidence potential pollution hotspots, providing a reliable spatial decision-making basis for subsequent targeted and precise monitoring and efficient intervention.

[0049] The air-ground collaborative positioning module, if the concentration in a potential pollution hotspot exceeds a preset risk threshold, immediately triggers a collaborative mechanism between the portable XRF detection device and the drone to acquire high-resolution soil texture images and rough heavy metal concentration data, and then guides the portable XRF detection device to locate a highly polluted and highly representative point.

[0050] Furthermore, the air-to-ground cooperative positioning module includes:

[0051] S11. If the concentration exceeds the preset risk threshold, the collaborative operation mechanism of the portable XRF detection device and the drone is triggered. The drone synchronously acquires high-resolution soil texture images of the hotspot area and uses its portable XRF preprocessing probe to scan the area to obtain the corresponding soil texture distribution details and rough heavy metal concentration distribution information.

[0052] Furthermore, based on the detailed distribution of soil texture and the approximate concentration distribution of heavy metals, the method also includes: accurately stitching multi-view soil texture images using SIFT feature points and the RANSAC algorithm; simultaneously using an extended Kalman filter to fuse GNSS / IMU navigation data to obtain centimeter-level positioning information with timestamps; unifying image pixel coordinates to the geodetic coordinate system through a coordinate transformation model to establish a spatiotemporal benchmark with a reprojection error better than 3 pixels; and finally, monitoring the registration residuals in real time through a quality control system and employing a pyramid hierarchical strategy to ensure a high degree of consistency between the texture image and the heavy metal concentration data in the spatiotemporal dimensions.

[0053] S12. Based on the texture distribution details and the concentration distribution information, a predetermined correlation analysis model is used to perform data fusion to generate a soil pollution distribution map, and the preliminary coordinates of the high pollution points are determined. The preliminary coordinates are sent to the positioning system of the portable XRF detection device to generate a path planning command, which drives the portable XRF detection device to move to the high pollution point to complete the precise positioning.

[0054] The predetermined correlation analysis model is a support vector machine model trained on historical data. The input features of the model are gray-level co-occurrence matrix features extracted from the texture image and various heavy metal concentration values ​​obtained from the XRF preprocessing probe; the output is the pollution probability score for each grid cell.

[0055] The precise positioning process involves the portable XRF detection device's positioning system receiving initial coordinates and then using Algorithm A for global path planning. This planning comprehensively considers the device's physical turning radius, known obstacle information, and terrain slope to generate a safe, efficient, and collision-free path. The portable XRF detection device's control system (typically an unmanned ground vehicle equipped with XRF sensors) follows the generated path instructions and, combined with centimeter-level position feedback provided by its real-time dynamic positioning system, autonomously navigates to the target location within a preset tolerance range. Upon reaching the predetermined location, the device's control system sends a positioning ready signal, completing the entire air-ground collaborative precise positioning process, and the system switches to a state of waiting for fine-tuning.

[0056] Furthermore, algorithm A is invoked for global path planning, specifically including:

[0057] Based on previously acquired high-resolution orthophotos and digital elevation models, a two-dimensional raster map for path planning is constructed. Each raster cell in the map is assigned a specific passage cost attribute. The assignment rules are as follows: flat and unobstructed bare soil areas are assigned a baseline cost of 1; vegetation-covered areas are assigned a cost of 2 due to increased passage resistance; and steep slopes with a gradient exceeding 15°, water bodies, and areas with known fixed obstacles (such as boulders and ditches) are assigned an infinite cost and marked as impassable areas. Finally, a raster cost map with different passage costs is generated, covering the target working area.

[0058] Using the current location of the portable XRF detection device as the starting point of the path and the constructed cost map as input, the A algorithm is executed to search for the optimal collision-free path to the target point. The evaluation function of the A algorithm is defined as f(n) = g(n) + h(n), where: g(n) is the actual cumulative cost from the starting point to the current grid n, which is calculated by combining the travel distance and the passage cost of the grids passed; h(n) is the heuristically predicted cost from the current grid n to the target point, which is calculated using the diagonal distance, which balances computational efficiency and accuracy. During the search process, collision detection is performed simultaneously to ensure that the generated path meets the minimum turning radius constraint of the device, that is, the radius of curvature of any point on the path is greater than the minimum turning radius of the device.

[0059] The initial shape of the obtained optimal path is a polyline formed by connecting the center points of the grid. To improve the stability and execution efficiency of vehicle tracking, a B-spline curve fitting algorithm is used to smooth the polyline path. By inserting control points, a final path with continuous curvature and smoothness is generated, and it is ensured that the smoothed path is still completely within the passable area and meets all physical constraints.

[0060] The smoothed final path is discretized into a series of dense waypoint sequences with latitude and longitude coordinates, and encapsulated into a path planning instruction data packet, which is then sent to the motion control system of the portable XRF detection device. At the same time, the system state switches to path ready, waiting to receive a positioning confirmation signal from the ground equipment to trigger subsequent fine detection tasks.

[0061] Specifically, after identifying potential pollution hotspots where concentrations exceed thresholds, an air-ground collaborative mechanism between UAVs and ground-based XRF detection equipment is immediately triggered. The UAVs simultaneously acquire high-resolution soil texture images and heavy metal concentration data. A precise spatiotemporal benchmark is established through image stitching, navigation data fusion, and coordinate unification. Then, a pollution distribution map is generated based on the support vector machine model fusion analysis, and preliminary coordinates are determined. Finally, the A-path planning algorithm guides the ground equipment to accurately reach the target location. This effectively solves the problems of missed or misjudged pollution points caused by low efficiency of manual sampling and inaccurate spatial positioning in traditional soil monitoring. It achieves a rapid and automated closed loop from regional screening to centimeter-level precise positioning of high-pollution points, greatly improving the efficiency and positioning accuracy of soil pollution monitoring.

[0062] The detection signal dynamic correction module uses a convolutional neural network (CNN)-random forest fusion model to dynamically correct interference in the XRF detection signal based on soil texture data and previous environmental parameters.

[0063] Furthermore, the detection signal dynamic correction module includes:

[0064] A fusion correction model consisting of a concatenated convolutional neural network and a random forest model is pre-established. The soil texture data is input into the convolutional neural network, which automatically performs deep feature learning to extract key environmental factors related to XRF detection signal interference. The key environmental factors include at least soil moisture gradient, texture variation characteristics, and organic matter distribution information.

[0065] Specifically, deep feature learning is automatically performed using convolutional neural networks to extract key environmental factors related to XRF detection signal interference. This includes: inputting real-time collected soil texture data (usually high-resolution RGB or multispectral images) into a pre-trained CNN model. The CNN model (preferably a simplified variant of VGG or ResNet) automatically performs deep feature learning through its multiple convolutional and pooling layers, extracting high-level key environmental factors related to XRF detection signal interference from pixel-level information. These factors are abstract features that cannot be directly observed using traditional methods, but systematically characterize at least the physicochemical properties of the soil, including moisture gradient, texture variation characteristics, and organic matter distribution information.

[0066] The key environmental factors extracted by the convolutional neural network are combined with the previous environmental parameters to form a fusion feature set, which is then input into the random forest model. The random forest model quantifies the interference weight of each environmental factor in the fusion feature set on the XRF detection signal, and outputs a comprehensive dynamic interference correction coefficient based on the interference weight.

[0067] Among them, the dynamic interference correction coefficient based on the comprehensive output of interference weights specifically includes: concatenating the key environmental factor feature vectors automatically extracted by the convolutional neural network with the environmental parameter data measured in the previous period to form a comprehensive fusion feature set; performing standardized preprocessing on the fusion feature set so that the numerical distribution of each feature satisfies the mean of zero and the standard deviation of one, so as to eliminate the impact of differences in the units and value ranges of different features on the model performance;

[0068] The standardized fusion feature set is input into a pre-generated random forest regression model, which consists of multiple decision trees trained in parallel. The calculation process of the interference weight is as follows: The system automatically counts the total frequency of each environmental factor being selected as a node splitting feature in all decision trees, and simultaneously calculates the amount of data impurity reduction caused by each split. Finally, the interference weight of each factor is obtained by normalizing the total amount of impurity reduction it contributes. This weight value directly quantifies the degree of influence of the factor on the XRF detection signal interference.

[0069] After obtaining the interference weights of each environmental factor, a comprehensive dynamic interference correction coefficient is generated through a preset weighting calculation rule. Specifically, the standardized eigenvalue of each factor is multiplied by its corresponding interference weight, and the weighted results of all factors are summed. A baseline bias term determined through model training is added, and finally a dimensionless scalar coefficient is output. The magnitude and sign of this coefficient fully characterize the comprehensive interference intensity and direction of the current soil environment on the XRF detection signal.

[0070] The calculated dynamic interference correction coefficients and their corresponding interference weight vectors are encapsulated together to form a structured correction data packet. This data packet is transmitted in real time to the pollution concentration calibration calculation module through the data interface defined within the system, providing authoritative parameter basis for subsequent accurate signal correction. This completes the core processing flow of dynamic interference correction.

[0071] Specifically, by constructing a cascaded correction model of convolutional neural network and random forest, the CNN automatically learns and extracts key environmental factors such as humidity gradient, texture variation and organic matter distribution from soil texture data. Then, the random forest model quantifies the interference weight of each environmental factor on the XRF detection signal, and finally generates dynamic interference correction coefficients. This effectively solves the problem of inaccurate measurement results caused by interference from complex soil environmental factors in traditional XRF detection methods, and significantly improves the anti-interference ability and detection accuracy of heavy metal characteristic spectral signals in complex field environments.

[0072] The pollution concentration calibration calculation module substitutes the dynamic interference correction coefficient into the initial environmental feature set and combines it with the heavy metal characteristic spectral line baseline correction using portable XRF technology to obtain a calibrated pollution concentration estimate that is close to the true value.

[0073] Furthermore, the pollution concentration calibration calculation module includes:

[0074] The dynamic interference correction coefficient is used as a weight to perform weighted correction on the environmental parameters in the initial environmental feature set, resulting in corrected environmental feature data that better reflects the real soil background.

[0075] The heavy metal characteristic spectra acquired by the XRF device were first denoised using the Savitzky-Golay convolution smoothing method, and then the smoothed spectra were corrected using an adaptive iterative baseline fitting algorithm to accurately extract the peak intensity and area characteristic values ​​of the spectral lines representing the heavy metal concentration.

[0076] The process involves several steps. First, Savitzky-Golay convolutional smoothing and noise reduction is performed on the original heavy metal characteristic spectra. This is achieved by using a sliding window with 11 data points and a third-order polynomial for local weighted least-squares fitting. This process preserves the true shape of the spectra while effectively suppressing high-frequency random noise, resulting in smoothed spectral data. Next, adaptive iterative baseline correction is performed. Using the minimum value of the smoothed spectra as the initial baseline, all spectral data points that meet the baseline conditions are identified and connected through iterative calculations. Low-intensity points outside the peak range are selected to gradually fit an adaptive baseline that closely matches the true base variation of the spectra. Finally, the peak intensity of the spectral lines is determined by identifying local maxima in the clean spectral lines after baseline correction. The net area of ​​the spectral peak and the baseline enclosed region is calculated using the trapezoidal integral method, which serves as two key characteristic values ​​representing the heavy metal concentration.

[0077] The corrected environmental characteristic data and spectral characteristic values ​​are input into a preset concentration calculation model to obtain a preliminary estimate of the pollution concentration. Then, based on the rules established by historical calibration data, the preliminary estimate is subjected to deviation judgment and secondary correction. Finally, a calibrated pollution concentration estimate that is close to the true value is output and stored in the database.

[0078] The process involves using corrected environmental characteristic data and spectral feature values ​​to call a pre-set concentration calculation model. This model is based on a large amount of historical calibration data, covering different soil types such as loam, sand, and clay, and different environmental conditions such as high humidity (>60%), low pH (<5.5), and high organic matter (>3%). It includes standard soil samples with known heavy metal concentrations, ranging from 0.1 mg / kg to 1000 mg / kg, to meet different pollution levels and correlate samples with corresponding XRF detection spectral features and environmental parameters. All samples have undergone data cleaning and normalization. The model employs a unified processing approach, incorporating a dedicated characteristic spectral line-concentration calibration curve for each target heavy metal. For example, for lead, the peak intensity of its 220.35 keV characteristic spectral line is used as the x-axis, and the known concentration of the standard sample is used as the y-axis, forming a calibration curve through linear fitting. For heavy metals with nonlinear correlations, such as mercury, curves are constructed through nonlinear fitting. Additionally, an environmental factor compensation algorithm is built-in, such as compensation based on soil moisture gradient: for every 5% increase in soil moisture, the corresponding peak intensity is multiplied by a compensation coefficient of 0.95 to offset the weakening effect of moisture on the signal.

[0079] Compensation based on organic matter distribution: For every 1% increase in organic matter content, the peak area is multiplied by a correction ratio of 1.02 to correct the signal deviation caused by the adsorption of heavy metals by organic matter. During model execution, the preprocessed and corrected environmental feature data is first fused with spectral feature values ​​through feature correlation. The environmental feature data is then substituted into the built-in environmental factor compensation algorithm to perform a secondary correction on the spectral feature values, eliminating trace environmental residual interference that may still exist after the initial dynamic interference correction, such as slight adsorption deviations in spectral signals in high organic matter regions. Next, the compensated spectral feature values ​​are precisely matched with the characteristic spectral line-concentration calibration curve of the corresponding target heavy metal in the model. Calculations are performed according to the preset fitting algorithm. Specifically, for heavy metals with a linear correlation between spectral lines and concentration (correlation coefficient R²>0.95), such as cadmium, a multiple linear regression algorithm is used to establish a linear equation for calculating the spectral feature values, environmental compensation parameters, and concentration. For heavy metals with a non-linear correlation (R²<0.95), such as mercury, a support vector regression algorithm is used (by mapping the data to a high-dimensional space through a kernel function to achieve non-linear fitting). Finally, a preliminary estimate of the target heavy metal pollution concentration at the detection point is output.

[0080] Specifically, by weighting environmental feature data with dynamic interference correction coefficients, combining Savitzky-Golay smoothing and adaptive baseline correction techniques to accurately extract spectral features, and using a concentration calculation model built on a large amount of historical data for environmental compensation and nonlinear fitting, the problem of concentration estimation bias caused by residual environmental interference and inaccurate spectral feature extraction in traditional XRF detection is solved. This achieves high-precision calibration of soil heavy metal pollution concentration, making the final result closer to the true value.

[0081] The sampling route optimization and planning module compares the calibrated pollution concentration estimate with the preset pollution standard, introduces a risk classification model, generates a sampling route optimization sequence driven by two factors: risk level and geographical accessibility, and outputs the optimal sampling and intervention points.

[0082] Furthermore, the sampling path optimization and planning module includes:

[0083] S21. Based on the calibrated pollution concentration estimate, compare it with the preset standard to identify areas exceeding the standard. Based on the degree of exceeding the standard, use a risk classification model to determine the risk level of each area.

[0084] The risk classification model is used to determine the risk level of each area. This includes receiving the calibrated pollution concentration estimate with geographic coordinates and calling the built-in national standard pollution risk screening value and control value library. The concentration of each heavy metal at each point is compared with its corresponding standard value one by one. If the concentration is equal to or higher than the risk screening value, the point is identified as an area exceeding the standard.

[0085] For each location exceeding the standard, calculate the single-factor pollution index for each heavy metal exceeding the standard. The calculation formula is heavy metal concentration / risk screening value. At the same time, implement the priority judgment rule: if the concentration of any heavy metal in the area exceeds the risk control value, the area is directly classified as extremely high risk level, regardless of other indicators.

[0086] For areas that do not reach extremely high risk, a comprehensive assessment is conducted from two dimensions: pollution intensity and pollution extent. Pollution intensity is quantified by calculating the Nemerow Comprehensive Pollution Index, which represents the largest pollution contribution, while pollution extent is assessed by counting the number of heavy metals exceeding the standard. Based on preset thresholds, areas with a Nemerow Index greater than 2.0 or with 3 or more exceeding the standard are classified as high-risk; areas with a Nemerow Index between 1.0 and 2.0 and with 1 to 2 exceeding the standard are classified as medium-risk.

[0087] Ultimately, a clear risk level label is output for each assessment area, including extremely high risk, high risk, medium risk, and low risk. This comprehensive risk level will serve as the core decision-making basis for subsequent sampling path optimization and governance plan recommendations, completing the transformation from concentration data to risk management.

[0088] S22. Integrate the risk level and geographical accessibility data to calculate a comprehensive priority score for each location to be planned; wherein, the geographical accessibility is quantified based on road network, slope and transportation cost data;

[0089] S23. Using the comprehensive priority score as the core input, a constrained path planning algorithm is used to generate a sampling and intervention path sequence with the highest overall efficiency that can prioritize the coverage of high-risk points.

[0090] Furthermore, the comprehensive priority scoring includes: constructing a scoring model using the analytic hierarchy process (AHP), establishing a risk level quantification matrix, and assigning values ​​of 8, 4, 2, and 1 to the four risk levels (extremely high risk to low risk) according to exponential differences to strengthen the decision-making weight of high-risk levels; simultaneously, constructing an accessibility assessment system based on multi-source geographic data: calculating the optimal path distance from a location to the nearest main road using the Dijkstra algorithm, extracting the slope cosine value using the digital elevation model to assess the difficulty of passage, setting differentiated passage cost coefficients based on land cover type, determining the weight of each indicator using the entropy weight method, and generating a 0-1 standardized accessibility score using the TOPSIS method; finally, using the weighted aggregation function S=0.7×(risk quantification value / 8)+0.3×accessibility score, and performing nonlinear calibration using the Sigmoid function, the final output is a comprehensive priority score in the 0-100 range, where a benchmark score threshold (≥60 points) is set for high-risk locations to ensure the risk-driven principle;

[0091] The constrained path planning algorithm includes: using an improved genetic algorithm to serialize the sampling points into chromosomes through integer encoding; designing a fitness function with comprehensive priority score coverage as the core and total path length and time window violation as penalty terms; employing tournament selection, sequential crossover, and reverse mutation operators during the evolution process, and embedding 2-opt local search to optimize path segments; and using hard constraints to ensure that the daily operation time is ≤8 hours and that high-risk points are fully covered; and finally outputting the optimal path sequence that balances risk coverage and operation efficiency based on the Pareto front.

[0092] Specifically, by comparing the calibrated pollution concentration with national standard values, a multi-level risk assessment system is constructed based on single-factor indices, Nemerow composite indices, and the number of types of heavy metals exceeding the standard. Geographical accessibility factors such as road network, slope, and transportation costs are integrated, and a comprehensive priority score is generated using a analytic hierarchy process-entropy weighted TOPSIS hybrid algorithm. Then, a constrained path optimization is performed using an improved genetic algorithm. This solves the problem in traditional soil sampling planning that only considers spatial distance while ignoring differences in pollution risk and on-site access conditions. It achieves intelligent path planning that synergistically optimizes risk priority and operational efficiency, ensuring priority coverage of high-risk areas and a significant improvement in overall operational efficiency.

[0093] The real-time remediation plan recommendation module uses portable XRF to accurately extract the characteristic spectral lines of target heavy metal ions for sampling and intervention sites (to confirm the type and purity of heavy metals). Combined with previous environmental data such as soil type and humidity, it calls up a differentiated remediation plan library to generate a real-time remediation recommendation sequence, and the recommendation results are associated with the risk level of the site.

[0094] Furthermore, the real-time recommendation module for the governance scheme includes:

[0095] Soil samples from sampling and intervention sites were analyzed using portable XRF equipment to obtain characteristic spectral lines of the target heavy metals and determine the specific types and concentrations of the heavy metals.

[0096] Specifically, the method uses a spectral deconvolution algorithm to separate overlapping peaks based on the characteristic spectral lines of the target heavy metals, determines the heavy metal species by peak energy location, calculates the element concentration based on the net peak area, and uses the Compton-Rayleigh ratio analysis method to assess the soil matrix effect, corrects the measured values, and outputs an accurate pollution inventory containing heavy metal species, concentrations, and uncertainties.

[0097] The heavy metal concentration is compared with a preset threshold, and combined with previously acquired soil type and moisture data, a final risk level label is assigned to each location;

[0098] Based on the types, concentrations, risk levels of heavy metals and soil environmental data, one or more of the most suitable remediation measures are matched from a pre-stored remediation solution library.

[0099] For the matched governance measures, they are prioritized according to processing efficiency, cost and the urgency of site risks, generating a customized real-time governance recommendation sequence for each site and outputting a structured governance solution report.

[0100] Specifically, by using portable XRF equipment to accurately analyze the characteristic spectral lines of heavy metals, combined with environmental parameters such as soil type and humidity, a complete pollution inventory including pollution type, concentration, and uncertainty is constructed. Based on the results of multi-dimensional risk assessment, differentiated remediation measures are intelligently matched from a pre-stored remediation solution library. Finally, a customized recommendation sequence is generated based on treatment efficiency, cost, and risk urgency. This solves the problems of traditional soil remediation relying on human experience, insufficient solution targeting, and low decision-making efficiency. It achieves a rapid closed loop from pollution detection to accurate recommendation of remediation solutions, significantly improving the scientific nature and timeliness of soil pollution remediation.

[0101] Example 2

[0102] This embodiment provides another portable XRF intelligent detection and early warning system for heavy metals in soil that integrates risk assessment functions. It transforms the monitoring and early warning system from a traditional auxiliary tool to an autonomous decision-making entity, reducing reliance on manual intervention while improving system response speed and handling accuracy. The differences from Embodiment 1 include:

[0103] The edge intelligent sensing module employs distributed, miniaturized XRF sensing nodes. Each node integrates a microelectromechanical system (MEMS) spectrometer and a low-power environmental sensor. Using a pulse sampling strategy, it maintains environmental parameter monitoring in standby mode, automatically activating XRF detection when abnormal fluctuations are detected. All nodes form a self-organizing network via the LoRaWAN protocol, enabling collaborative monitoring and data forwarding across wide areas.

[0104] The digital twin modeling module constructs a 3D geological model of the target area based on BIM+GIS technology. After fusing real-time monitoring data with historical data, it simulates the migration and diffusion patterns of heavy metals in the soil through finite element analysis. A dynamic data-driven mechanism is employed, automatically calibrating model parameters every 24 hours to ensure the synchronization between the digital twin and the physical entity.

[0105] The blockchain evidence storage module utilizes lightweight blockchain technology to generate a cryptographic hash value containing a timestamp, geographical location, and device ID for each block of detection data. Through smart contracts, it achieves immutable storage of detection data, establishes a distributed ledger shared by multiple parties, and provides credible electronic evidence for environmental law enforcement and pollution liability determination.

[0106] The adaptive early warning module constructs an early warning strategy optimization model based on deep reinforcement learning algorithms, and continuously adjusts the early warning threshold and response strategy through multi-agent collaborative training. The system can dynamically optimize the early warning rule base based on regional characteristics, seasonal changes, and historical event data, realizing the transformation from fixed threshold early warning to intelligent strategy early warning.

[0107] The augmented reality interaction module overlays pollution distribution, risk assessment results, and remediation plans onto the real scene in a holographic projection through terminal devices such as AR glasses. It supports a virtual operation interface with gesture interaction and can realize functions such as 3D annotation of polluted areas and simulation demonstration of remediation plans, which greatly improves the intuitiveness and operational efficiency of on-site operations.

[0108] The autonomous decision-making and governance module integrates control interfaces for various automated remediation equipment, automatically dispatches corresponding remediation resources according to the warning level, and presets multiple treatment plans for different pollution scenarios, including precise spraying of remediation agents by drones and automatic deployment of permeable reactive barriers, thus realizing a complete closed loop from monitoring and early warning to autonomous governance.

[0109] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An integrated risk assessment function of soil heavy metal portable XRF intelligent detection and early warning system, characterized in that: The system comprises a data acquisition and integration module, a pollution hotspot area identification module, an air-ground collaborative positioning module, a detection signal dynamic correction module, a pollution concentration calibration calculation module, a sampling path optimization planning module, and a treatment scheme real-time recommendation module. The data acquisition and integration module acquires soil heavy metal characteristic spectral data and multi-dimensional environmental parameters in the target area by deploying a portable XRF front-end module integrated with a multi-spectral sensor and a miniature environmental probe, and forms an initial environmental feature set in two dimensions of spectrum and environment. The pollution hotspot area identification module uses an improved density clustering algorithm to determine high-confidence potential pollution hotspot areas based on the initial environmental feature set and by integrating regional prior knowledge and associating heavy metal pollution probability while grouping soil types. The air-ground collaborative positioning module triggers a portable XRF detection device and a UAV collaborative mechanism if the concentration of the potential pollution hotspot area exceeds a preset risk threshold, acquires high-resolution soil texture images and rough heavy metal concentration data, and then guides the portable XRF detection device to position to a high-pollution-high-representative point. The detection signal dynamic correction module uses a convolutional neural network-random forest fusion model to perform dynamic interference correction on the XRF detection signal based on soil texture data and previous environmental parameters. The pollution concentration calibration calculation module substitutes the dynamic interference correction coefficient into the initial environmental feature set, combines the baseline correction of the heavy metal characteristic spectral line of the portable XRF technology, and obtains calibrated pollution concentration estimates close to the true values. The sampling path optimization planning module compares the calibrated pollution concentration estimates with the preset pollution standard, introduces a risk classification model, generates a sampling path optimization sequence driven by risk levels and geographical accessibility, and outputs optimal sampling and intervention points. The treatment scheme real-time recommendation module extracts target heavy metal ion characteristic spectral lines accurately based on the sampling and intervention points, combines previous environmental data, calls a differentiated treatment scheme library, generates a real-time treatment recommendation sequence, and associates the recommendation results with point risk levels. 2.The soil heavy metal portable XRF intelligent detection and early warning system integrated with risk assessment function according to claim 1, characterized in that: The data acquisition and integration module comprises: The portable XRF front-end module is started, and the heavy metal characteristic spectral signals of the soil are synchronously excited and collected in the target area according to a preset grid point, and at the same time, the multi-dimensional environmental parameters including soil humidity, pH value, and temperature are recorded by the miniature environmental probe. The original spectral data and environmental parameter data collected are respectively standardized in format and dimension, and the wavelet transform algorithm is used for noise filtering to obtain clean spectral data set and environmental data set. If there are missing values in the preprocessed data set, the kriging spatial interpolation method is used for completion to form a complete spatial data set. The principal component analysis method is used for dimension reduction processing of the supplemented complete data, and the key principal components with a contribution rate exceeding a preset threshold are extracted to form a simplified feature combination. Based on the simplified feature combination, a structured initial environmental feature set is constructed, which represents the spatial distribution characteristics of soil heavy metals in the target area, and is finally compressed and encrypted and stored in a preset cloud or local database. 3.The soil heavy metal portable XRF intelligent detection and early warning system integrated with risk assessment function according to claim 1, characterized in that: The pollution hotspot area identification module comprises: The pre-stored regional soil type distribution map and historical pollution records are called and spatially associated with the initial environmental feature set to construct a multi-dimensional prior information library; an improved density clustering algorithm is used to cluster the spatial points by introducing a soil type constraint condition to optimize the distance metric, divide the soil class groups, and simultaneously calculate the average pollution probability of each cluster; A preliminary pollution risk distribution map is generated based on the clustering results; data outliers are filtered and removed by a preset outlier detection threshold to obtain a corrected risk distribution data set; a high confidence standard is set to sort potential pollution hot spot areas and generate a priority list; The priority list is visualized to generate a regional distribution map, and the soil type area boundary is matched with the potential pollution hot spot area using spatial overlay analysis tools in geographic information systems to identify high-overlap key focus area combinations; A pollution screening method is applied to the overlapping areas based on the principle of heavy metal synergistic effect to finally lock the high-risk pollution area set and output its geographic boundary coordinates.

4. The portable XRF intelligent detection and early warning system for soil heavy metals integrated with risk assessment function according to claim 3, characterized in that: The improved density clustering algorithm considers the influence of soil type difference on clustering results by introducing an improved distance measurement function, which is specifically expressed as: ; is the Euclidean distance between point i and point j; is an indicator function, which takes the value of 1 when the soil types of point i and point j are inconsistent, and 0 otherwise; is a type penalty factor, which is an adjustable parameter greater than 0, used to control the degree of influence of soil type difference on clustering results.

5. The portable XRF intelligent detection and early warning system for soil heavy metals integrated with risk assessment function according to claim 1, characterized in that: The air-ground cooperative positioning module comprises: S11, if the concentration exceeds the preset risk threshold, the cooperative working mechanism of the portable XRF detection device and the unmanned aerial vehicle is triggered, the high-resolution soil texture image of the hot spot area is synchronously obtained through the unmanned aerial vehicle, and the portable XRF pretreatment probe carried by the unmanned aerial vehicle is used to scan the area to obtain the corresponding soil texture distribution details and heavy metal rough concentration distribution information; S12, based on the texture distribution details and the concentration distribution information, a predetermined correlation analysis model is used for data fusion to generate a soil pollution distribution map and determine the preliminary coordinates of the high pollution point, and the preliminary coordinates are sent to the positioning system of the portable XRF detection device to generate path planning instructions to drive the portable XRF detection device to move to the high pollution point to complete positioning.

6. The portable XRF intelligent detection and early warning system for soil heavy metals integrated with risk assessment function according to claim 5, characterized in that: After determining the preliminary coordinates of the high pollution point, it further comprises: generating path planning instructions by calling A algorithm for global path planning.

7. The portable XRF intelligent detection and early warning system for soil heavy metals integrated with risk assessment function according to claim 1, characterized in that: The detection signal dynamic correction module comprises: A fusion correction model composed of a convolutional neural network and a random forest model is established in advance, the soil texture data is input into the convolutional neural network, the convolutional neural network automatically learns deep features, and key environmental factors related to XRF detection signal interference are extracted; the key environmental factors at least include soil humidity gradient, texture variation characteristics, and organic matter distribution information; The key environmental factors extracted by the convolutional neural network and the previous environmental parameters form a fusion feature set, which is input into the random forest model; the random forest model quantifies the interference weight of each environmental factor in the fusion feature set on the XRF detection signal, and outputs a comprehensive dynamic interference correction coefficient based on the interference weight.

8. The portable XRF intelligent detection and early warning system for soil heavy metals integrated with risk assessment function according to claim 1, characterized in that: The pollution concentration calibration calculation module comprises: The dynamic interference correction coefficient is used as a weight to modify the environmental parameters in the initial environmental feature set to obtain modified environmental feature data that better reflects the real soil background; The characteristic spectral line of the heavy metal collected by the XRF device is first denoised by using the Savitzky-Golay convolution smoothing method, and then the baseline of the smoothed spectral line is corrected by using the adaptive iterative baseline fitting algorithm, so as to accurately extract the spectral line peak intensity and area characteristic value representing the concentration of the heavy metal; The modified environmental feature data and the spectral line characteristic value are input into a preset concentration calculation model to obtain a preliminary estimated value of the pollution concentration; the preliminary estimated value is subjected to deviation judgment and secondary correction according to a rule established based on historical calibration data, and finally an estimated result of the calibrated pollution concentration close to the true value is output and stored in a database.

9. The portable XRF intelligent detection and early warning system for soil heavy metals integrated with risk assessment function according to claim 1, characterized in that: The sampling path optimization planning module comprises: S21, comparing the calibrated pollution concentration estimated value with a preset standard to identify an over-standard area, and based on the over-standard degree, using a risk grading model to determine a risk level for each area; S22, fusing the risk level and geographic accessibility data to calculate a comprehensive priority score for each to-be-planned point; wherein the geographic accessibility is quantified based on road network, slope and transportation cost data; S23, using a path planning algorithm with constraints to generate a sampling and intervention path sequence with the highest overall efficiency and preferentially covering high-risk point locations, with the comprehensive priority score as the core input.

10. The portable XRF intelligent detection and early warning system for soil heavy metals integrated with risk assessment function according to claim 1, characterized in that: The governance scheme real-time recommendation module comprises: detecting the soil samples of the sampling points and intervention points by using the portable XRF device to obtain the characteristic spectral line of the target heavy metal and determine the specific type and concentration of the heavy metal; comparing the heavy metal concentration with a preset threshold value, and combining the soil type and humidity data obtained in the early stage to assign a final risk level label to each point; based on the type, concentration and risk level of the heavy metal and the soil environment data, matching one or more most suitable governance measures from a pre-stored governance scheme library; for the matched governance measures, performing priority sorting according to the processing efficiency, cost and urgency of the point risk to generate a customized real-time governance recommendation sequence for each point, and outputting a structured governance scheme report.

Images