Soil heavy metal pollution intelligent identification and dynamic early warning system and method

Abstract

The invention relates to the technical field of environment monitoring, in particular to a soil heavy metal pollution intelligent identification and dynamic early warning system and method, and the system comprises a data collection and preprocessing module which is used for obtaining multi-source satellite hyperspectral data of a to-be-detected region, carrying out the data preprocessing and data fusion of the data, and obtaining the surface reflectance data of the to-be-detected region; the spectral feature enhancement and extraction module is used for carrying out soil pixel purification and spectral feature enhancement on the surface reflectance data and extracting sensitive spectral features; the data analysis and inversion module is used for inputting the sensitive spectral characteristics into a preset intelligent inversion and early warning model library to obtain a soil heavy metal pollution grade identification result and a quantitative concentration inversion result; and the dynamic early warning module is used for carrying out time-space diffusion trend analysis and dynamic risk assessment on soil heavy metal pollution and generating a risk early warning report. The soil heavy metal pollution monitoring method is low in cost, high in efficiency and wide in range, and a soil heavy metal pollution monitoring solution of business operation can be realized.

Description

Technical Field

[0001] This application relates to the field of environmental monitoring technology, and in particular to an intelligent identification and dynamic early warning system and method for heavy metal pollution in soil. Background Technology

[0002] Heavy metal pollution in soil is a significant issue in environmental monitoring and food safety. Currently, monitoring of heavy metals in soil mainly relies on traditional ground sampling and laboratory analysis methods. While these methods offer high analytical accuracy, they are costly and time-consuming due to the need for manual sampling, collection, and laboratory processing. Furthermore, the sparse sampling points make it difficult to achieve large-scale, continuous spatial monitoring.

[0003] To expand monitoring coverage, airborne hyperspectral remote sensing technology has been applied to regional environmental monitoring. This technology boasts high spectral resolution, enabling the identification of subtle spectral features of ground objects. However, due to the high operating costs of airborne platforms and significant limitations imposed by airspace and weather conditions, it is difficult to achieve frequent, operational monitoring tasks.

[0004] In satellite remote sensing, multispectral satellite data (such as Landset and Sentinel-2) has been attempted for soil environmental monitoring. However, due to its limited number of bands and low spectral resolution, it cannot effectively capture the weak spectral characteristics caused by heavy metals, and therefore is mainly used for qualitative or semi-quantitative analysis, making it difficult to achieve accurate concentration inversion. Although early hyperspectral satellites can provide richer spectral information, due to the difficulty in data acquisition, complex preprocessing, and lack of robust inversion models, a mature operational application system has not yet been formed.

[0005] Furthermore, existing remote sensing monitoring of heavy metals in soil mainly focuses on identification and retrieval, lacking the ability to dynamically predict pollution trends. Due to the failure to effectively integrate multi-temporal data and geographic auxiliary information, existing technologies struggle to achieve pollution diffusion analysis and risk level assessment, thus limiting their decision support role in environmental risk prevention and control.

[0006] Therefore, there is an urgent need for an intelligent identification and dynamic early warning system and method for soil heavy metal pollution. Summary of the Invention

[0007] (a) Technical problems to be solved

[0008] In view of the above-mentioned problems and shortcomings of the prior art, this application provides a soil heavy metal pollution intelligent identification and dynamic early warning system and method, which solves the technical problems of prominent contradiction between monitoring range and cost, insufficient identification and inversion accuracy, and lack of effective dynamic early warning capability in the prior art.

[0009] (II) Technical Solution

[0010] To achieve the above objectives, the main technical solutions adopted in this application include:

[0011] In a first aspect, embodiments of this application provide an intelligent identification and dynamic early warning system for soil heavy metal pollution, comprising:

[0012] The data acquisition and preprocessing module is used to acquire multi-source satellite hyperspectral data of the area to be measured, and to perform data preprocessing and data fusion on the multi-source satellite hyperspectral data to obtain the surface reflectance data of the area to be measured.

[0013] The spectral feature enhancement and extraction module is used to purify soil pixels and enhance spectral features of the surface reflectance data of the area to be tested, and extract sensitive spectral features related to soil heavy metal content.

[0014] The data analysis and inversion module is used to input the sensitive spectral features into a preset intelligent inversion and early warning model library to obtain soil heavy metal pollution level identification results and quantitative concentration inversion results;

[0015] The dynamic early warning module is used to analyze the spatiotemporal diffusion trend of soil heavy metal pollution and conduct dynamic risk assessment based on the soil heavy metal pollution level identification results and quantitative concentration inversion results, combined with multi-temporal satellite data and environmental auxiliary geographic data covering the area to be tested, and generate a risk early warning report.

[0016] Optionally, in some embodiments of this application, the data acquisition and preprocessing module includes:

[0017] A radiometric calibration unit is used to convert the multi-source satellite hyperspectral data into apparent radiance data;

[0018] An atmospheric correction unit is used to process the apparent radiance data using a 6S or MODTRAN model to eliminate atmospheric effects and obtain initial surface reflectance data.

[0019] The geometric correction unit is used to perform geometric fine correction on the initial surface reflectance data based on ground control points and digital elevation model, and obtain the geometrically corrected surface reflectance data.

[0020] The data fusion unit is used to perform spatial-spectral fusion of the geometrically corrected surface reflectance data with the simultaneous multispectral data to obtain the surface reflectance data of the area to be measured.

[0021] Optionally, in some embodiments of this application, the spectral feature enhancement and extraction module includes:

[0022] The pure pixel extraction unit is used to classify the surface reflectance data of the area to be measured using vegetation index and water index, and remove pixels in vegetation-covered and water-covered areas by setting a threshold mask to extract the surface reflectance data of pure soil pixels.

[0023] The spectral enhancement unit is used to perform at least one transformation on the surface reflectance data of the pure soil pixels, including spectral differentiation, continuum removal, or absorption depth analysis, to obtain spectrally enhanced surface reflectance data.

[0024] The feature selection unit is used to use a genetic algorithm to screen out sensitive spectral features related to soil heavy metal content from spectrally enhanced surface reflectance data.

[0025] Optionally, in some embodiments of this application, the feature selection unit is specifically used for:

[0026] The full-band set of the spectrally enhanced surface reflectance data is used as the initial population.

[0027] The fitness value of an individual in the population is obtained by using the prediction accuracy of the heavy metal concentration inversion model constructed based on the selected band combination as the fitness function.

[0028] Based on the fitness value, an individual is selected to enter the next generation using roulette wheel selection.

[0029] Perform crossover and mutation operations on the selected individuals with preset probabilities to generate a new offspring population;

[0030] The selection, crossover, and mutation operations are performed iteratively until the preset number of generations is reached, and the combination of bands represented by the individuals with the highest fitness in each generation is used as the final selected sensitive spectral features.

[0031] Optionally, in some embodiments of this application, the data analysis and inversion module includes:

[0032] The initial inversion unit is used to simultaneously input the sensitive spectral features into the pollution level identification model and the quantitative concentration inversion model to obtain the initial pollution level and the initial quantitative concentration, respectively.

[0033] The bidirectional feedback unit is used to map the initial quantitative concentration value to a predefined concentration range label, and to input the concentration range label as a new feature dimension, together with the sensitive spectral feature, into the pollution level identification model for secondary discrimination, so as to output the final soil heavy metal pollution level identification result; it is also used to use the preset concentration threshold range corresponding to the initial pollution level as the constraint condition of the quantitative concentration inversion model, and to correct the initial quantitative concentration value that exceeds the range, so as to constrain it within the threshold range, so as to obtain the final quantitative concentration inversion result.

[0034] Optionally, in some embodiments of this application, the pollution level identification model is a multi-classification model based on machine learning, and the quantitative concentration inversion model is a regression model based on machine learning;

[0035] The pollution level identification model and the quantitative concentration inversion model are jointly trained using a dual-task learning framework that shares a low-level feature network. The loss function during training is a weighted sum of regression loss and classification loss.

[0036] Optionally, in some embodiments of this application, the dynamic early warning module includes:

[0037] The spatiotemporal trend analysis unit is used to simulate the spatiotemporal diffusion trend of soil heavy metal pollution based on the quantitative concentration inversion results of multiple time phases, using a cellular automata model or a geographically weighted regression model.

[0038] The dynamic risk assessment unit is used to integrate the pollution diffusion trend, pollution level identification results and environmental auxiliary geographic data to conduct pollutant migration simulation and risk assessment, and to delineate dynamic risk level areas.

[0039] The report generation unit is used to generate a risk warning report containing the spatial distribution of risks and key risk sources based on the pollution diffusion trend and dynamic risk level area.

[0040] Optionally, in some embodiments of this application, the dynamic risk assessment unit includes:

[0041] The spatiotemporal sequence construction subunit is used to construct a pixel-level time series dataset of heavy metal concentration and pollution level based on the quantitative concentration inversion results and pollution level identification results of multiple time phases.

[0042] The early warning analysis subunit is used to analyze pollution trends based on the time series dataset using time series forecasting, and to perform diffusion simulations in conjunction with environmental auxiliary geographic data, thereby generating trend-based early warning and diffusion-based early warning information.

[0043] Optionally, in some embodiments of this application, the early warning analysis subunit includes:

[0044] The trend warning analysis node is used to fit the time series dataset using an autoregressive integral moving average model or a long short-term memory network model, predict the trend of pollution concentration changes in a specific future period, and issue trend warning signals for areas that exceed the pollution risk level threshold.

[0045] The diffusion early warning analysis node is used to combine water flow direction and cumulative volume data derived from the digital elevation model, as well as regional prevailing wind direction frequency data, to simulate the spatial diffusion path and impact range of pollutants based on the heavy metal migration model, and generate diffusion early warning signals.

[0046] Secondly, embodiments of this application provide a method for soil heavy metal pollution identification and dynamic early warning, including:

[0047] Multi-source satellite hyperspectral data of the area to be measured are acquired, and the multi-source satellite hyperspectral data are preprocessed and fused to obtain the surface reflectance data of the area to be measured.

[0048] Soil pixel purification and spectral feature enhancement were performed on the surface reflectance data of the area to be tested to extract sensitive spectral features related to soil heavy metal content.

[0049] The sensitive spectral features are input into a preset intelligent inversion and early warning model library to obtain soil heavy metal pollution level identification results and quantitative concentration inversion results;

[0050] Based on the soil heavy metal pollution level identification results and quantitative concentration inversion results, combined with multi-temporal satellite data and environmental auxiliary geographic data covering the area to be tested, the spatiotemporal diffusion trend analysis and dynamic risk assessment of soil heavy metal pollution are carried out, and a risk warning report is generated.

[0051] (III) Beneficial Effects

[0052] The beneficial effects of this application are as follows: The intelligent identification and dynamic early warning system and method for soil heavy metal pollution of this application, by employing multi-source satellite hyperspectral data preprocessing and fusion technology, soil pixel purification and sensitive spectral feature enhancement extraction technology, intelligent inversion and early warning model library data analysis and inversion technology, and combining multi-temporal satellite data and environmental auxiliary geographic data spatiotemporal diffusion trend analysis and dynamic risk assessment technology, compared with the existing technology, can accurately realize the intelligent identification of soil heavy metal pollution level, efficient inversion of quantitative concentration, and dynamic tracking and risk early warning of pollution spatiotemporal diffusion trend. It achieves the technical effect of improving the intelligence level, accuracy and timeliness of soil heavy metal pollution monitoring and early warning, and providing scientific and reliable technical support for soil heavy metal pollution prevention and control and governance decision-making. Attached Figure Description

[0053] Figure 1 This is a structural block diagram of a soil heavy metal pollution intelligent identification and dynamic early warning system according to an embodiment of this application;

[0054] Figure 2 This is a flowchart illustrating a method for intelligent identification and dynamic early warning of heavy metal pollution in soil according to an embodiment of this application;

[0055] Figure 3 This is an example image of hyperspectral data for a method for intelligent identification and dynamic early warning of heavy metal pollution in soil according to an embodiment of this application. Detailed Implementation

[0056] To better explain and facilitate understanding of this application, the following detailed description of the application is provided in conjunction with the accompanying drawings and specific embodiments.

[0057] Among related technologies, intelligent identification and dynamic early warning of heavy metal pollution in soil can be summarized into the following categories:

[0058] The first category is traditional ground sampling and laboratory analysis methods. Although they have the advantage of high precision, they have problems such as high cost, long time consumption, and sparse monitoring points. They cannot reflect the spatial continuity and regional distribution pattern of pollution, and it is difficult to achieve large-scale rapid surveys.

[0059] The second category is airborne hyperspectral detection technology. Although it can cover a wide area, it has high flight costs, complex airspace application, and long monitoring cycle, which cannot meet the needs of routine and operational monitoring.

[0060] The third category is existing satellite remote sensing technology, but it has obvious limitations: multispectral satellites have wide bands and few numbers, making it difficult to capture subtle spectral features caused by heavy metals, and can only achieve qualitative or semi-quantitative analysis, with insufficient inversion accuracy; early hyperspectral satellite data acquisition was difficult and the preprocessing process was complex, and existing technologies mostly focus on pollution monitoring and concentration inversion, lacking the ability to dynamically track the spatiotemporal diffusion trend of pollution, failing to effectively transform monitoring data into predictive risk warning information, and unable to support early prevention and control decisions.

[0061] With the networking and operation of domestically produced Gaofen-5 (GF-5), Environmental Disaster Reduction-2 (HJ-2A / B), and international PRISMA and EnMAP, the ability to acquire hyperspectral data has been greatly improved. However, how to make full use of massive multi-source hyperspectral data, overcome the limitations of existing technologies in monitoring accuracy, automation, and dynamic early warning capabilities, and build an automated and operational intelligent identification and dynamic early warning system for soil heavy metal pollution has become a core technical problem that urgently needs to be solved.

[0062] To address the aforementioned technical issues, this application provides an intelligent identification and dynamic early warning system and method for soil heavy metal pollution. This system integrates satellite hyperspectral remote sensing, artificial intelligence, and GIS technologies. It works collaboratively through four modules: data acquisition and preprocessing, spectral feature enhancement and extraction, data analysis and inversion, and dynamic early warning. First, it acquires and optimizes multi-source hyperspectral data, extracting sensitive spectral features of heavy metals. Then, through dual-task learning and bidirectional feedback, it achieves high-precision inversion of pollution levels and concentrations. Finally, by combining multi-temporal data and environmental geographic information, it simulates the spatiotemporal diffusion trend of pollution and generates tiered early warning reports.

[0063] Through the above-mentioned technical solution, this application can achieve large-scale, low-cost operational monitoring, significantly improve the accuracy of pollution identification and concentration inversion, and at the same time make up for the shortcomings of dynamic early warning, providing efficient and reliable technical support for pollution prevention and control decision-making.

[0064] To better understand the above technical solutions, exemplary embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application can be understood more clearly and thoroughly, and that the scope of this application can be fully conveyed to those skilled in the art.

[0065] Example 1

[0066] Figure 1 This application describes an intelligent identification and dynamic early warning system for heavy metal pollution in soil, according to one embodiment. Figure 1 As shown, the system includes:

[0067] The data acquisition and preprocessing module is used to acquire multi-source satellite hyperspectral data of the area to be measured (hyperspectral data such as...). Figure 3 As shown in the figure, data preprocessing and data fusion were performed on multi-source satellite hyperspectral data to obtain the surface reflectance data of the area to be measured.

[0068] Specifically, the data acquisition and preprocessing module includes:

[0069] The radiometric calibration unit, the starting point of the entire processing flow, is crucial in converting multi-source satellite hyperspectral data (DN values) into apparent radiance data. It plays a key role in transforming the raw digital quantization values ​​recorded by the satellites into apparent radiance data with clear physical meaning. Multi-source satellite hyperspectral data is raw data transmitted from satellites; its essence is dimensionless electronic signals, influenced by various factors such as sensor response characteristics, lighting conditions, and observation angle. Radiometric calibration involves establishing a precise mathematical conversion model using calibration coefficients attached to the satellite data, converting DN values ​​into apparent radiance expressed in standard physical units.

[0070] The radiometric calibration unit eliminates the systematic errors of the sensor itself, enabling the data to accurately reflect the radiation energy intensity of ground objects at the sensor's entrance pupil, thus laying the foundation for subsequent atmospheric correction.

[0071] The Atmospheric Correction Unit (ACU) processes apparent radiance data using 6S or MODTRAN models to eliminate atmospheric effects and obtain initial surface reflectance data. Specifically, apparent radiance is the combined result of atmospheric path radiation and surface reflected radiation after atmospheric attenuation, containing a significant amount of non-surface information. The ACU employs widely validated atmospheric radiation models, such as 6S or MODTRAN models, to simulate the light transmission process in the atmosphere-surface system. The system takes into account atmospheric parameters such as transit time, geographic location, aerosol type and concentration, and water vapor content, and the model accurately calculates and subtracts the effects of atmospheric absorption and scattering. After processing by the ACU, initial surface reflectance data that directly reflects the spectral characteristics of the surface features is obtained, a prerequisite for comparative analysis between data from different times and different satellites.

[0072] The geometric correction unit is used to perform geometric fine correction on the initial surface reflectance data based on ground control points and digital elevation models, and to obtain the geometrically corrected surface reflectance data.

[0073] The core function of the geometric correction unit is to correct geometric distortions caused by factors such as sensor attitude changes, orbital offsets, and terrain undulations during satellite imaging. Based on ground control points and a digital elevation model (DEM), it performs precise geometric correction to ensure the spatial accuracy of surface reflectance data, meeting the needs of subsequent data fusion and spatial analysis. Geometric distortions in satellite data mainly include systematic distortions (caused by sensor design parameters and orbital parameters, such as panoramic distortion and scanline tilt) and non-systematic distortions (caused by attitude changes and terrain undulations, such as projection distortion and terrain displacement). Geometric correction requires establishing a mapping relationship between the original image and the geographic reference coordinate system using ground control points, and then using the DEM to complete terrain correction, eliminating pixel displacement errors caused by terrain undulations.

[0074] Specifically, the geometric correction unit uses ground control points as coordinate references and a digital elevation model as terrain basis. Through a specific correction model, it establishes a precise mapping relationship between the original image pixels and the real geographic coordinates, ultimately completing geometric precision correction. In key processes, ground control points need to be easily identifiable and locationally stable feature points, such as road intersections, bridge endpoints, reservoir dam corners, and surface building vertices. The number of control points needs to be evenly distributed throughout the entire image area, avoiding concentration in edge areas, and the planar position error of the control points needs to be ≤1 pixel. Furthermore, regarding the correction model, for satellite data with complete orbital parameters, such as the Sentinel series and GF series, a rational function model can be used. This model does not rely on sensor physical parameters and directly establishes the mapping relationship between image coordinates and geographic coordinates through control points, adapting to complex geometric distortions. For data sources with missing or simplified orbital parameters, a polynomial model can be used to calculate the geographic coordinates of pixels by fitting the distortion law through control points. Terrain correction uses the orthorectification method, combining digital elevation models and imaging geometric parameters to project the pixels of the tilted imaging onto the horizontal ground, eliminating terrain displacement and obtaining geometrically corrected surface reflectance data.

[0075] Furthermore, if the calibration model is a multi-image model, then the number of ground control points must be ≥6; if the calibration model is a rational function model, then the number of ground control points must be ≥12.

[0076] The data fusion unit is used to spatially and spectrally fuse geometrically corrected surface reflectance data with concurrent multispectral data to obtain surface reflectance data of the area to be measured.

[0077] In practical implementation, surface reflectance data converted from multi-source satellite hyperspectral data (i.e., geometrically corrected surface reflectance data) suffers from a disadvantage: while hyperspectral data boasts continuous and fine spectral resolution, its spatial resolution is often relatively low; conversely, multispectral data from the same phase typically possesses higher spatial resolution. Therefore, the data fusion unit employs advanced spatial-spectral fusion algorithms such as Gram-Schmidt, HySure, or deep learning-based algorithms to organically combine the rich spectral information of hyperspectral data with the clear spatial details of multispectral data. Essentially, it injects high-frequency spatial information (texture, edges, etc.) from high-spatial-resolution imagery into low-spatial-resolution hyperspectral imagery, ultimately generating a set of surface reflectance data that combines both high spectral and high spatial resolution.

[0078] In summary, the data acquisition and preprocessing module, through four precisely interconnected units—radiographic calibration, atmospheric correction, geometric correction, and data fusion—successfully transformed raw multi-source satellite hyperspectral data into high-quality, high-spatial-resolution surface reflectance data that can be directly used for quantitative analysis, providing reliable data support for the subsequent applications of the entire system.

[0079] The intelligent identification and dynamic early warning system for soil heavy metal pollution in this application also includes:

[0080] The spectral feature enhancement and extraction module is used to purify soil pixels and enhance spectral features of the surface reflectance data of the area to be tested, and extract sensitive spectral features related to soil heavy metal content.

[0081] The spectral feature enhancement and extraction module includes:

[0082] The pure soil pixel extraction unit is used to classify pixels using the vegetation index and water index of the land surface reflectance data of the area to be measured, and remove pixels in the vegetation cover area and water cover area by setting a threshold mask to extract the land surface reflectance data of pure soil pixels.

[0083] For vegetation indices, indices sensitive to vegetation cover and resistant to atmospheric interference are prioritized, including the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI). These two indices, used in conjunction, can cover all scenarios from low to high vegetation cover: NDVI is suitable for identifying areas with medium to high vegetation cover, while EVI, due to the introduction of an atmospheric correction term, is more suitable for accurately distinguishing low vegetation cover areas. For water body indices, the Normalized Difference Water Index (NDWI) and the Modified Normalized Difference Water Index (MNDWI), which are sensitive to differences between water and soil, are used. MNDWI, by replacing the shortwave infrared band, can effectively eliminate interference from soil background and buildings on water body identification, improving the identification accuracy of shallow and narrow water bodies.

[0084] Furthermore, based on the corresponding bands of high spatial resolution surface reflectance data, the four indices NDVI, EVI, NDWI, and MNDWI are calculated pixel by pixel to generate thematic maps of the four indices. The classification thresholds are set by combining "statistical analysis + visual verification": by statistically analyzing the frequency distribution histogram of index values ​​in the study area, the NDVI / EVI boundary threshold and the NDWI / MNDWI boundary threshold between vegetation and soil are determined. For example, NDVI ≥ 0.1 and EVI ≥ 0.05 can be identified as vegetation-covered areas, and similarly, NDWI ≥ 0.2 and MNDWI ≥ 0.15 can be identified as water-covered areas. Furthermore, vegetation and water masks are constructed based on set thresholds. "Reverse masking" is performed on the high spatial resolution surface reflectance data, that is, all pixels with NDVI / EVI exceeding the vegetation threshold and NDWI / MNDWI exceeding the water threshold are removed, and the remaining pixels are retained. For some mixed pixels, such as sparse vegetation areas with vegetation coverage of <5% and transition areas at the interface between soil and water, a linear mixture decomposition model is used for secondary purification. It is assumed that the spectrum of the mixed pixel is a linear mixture of the spectrum of soil and vegetation / water endmembers. The pure spectral contribution of soil endmembers is separated by the unmixing equation, and finally, high-purity pure soil pixel surface reflectance data is output.

[0085] The spectral enhancement unit is used to perform at least one transformation on the surface reflectance data of pure soil pixels, including spectral differentiation, continuum removal, or absorption depth analysis, to obtain spectrally enhanced surface reflectance data.

[0086] Spectral differentiation eliminates the linear or low-order polynomial trend of soil background reflectance by calculating the first and second derivatives of the spectral curve, highlighting the edge features of spectral absorption valleys while suppressing low-frequency noise. Specifically, the Savitzky-Golay smoothing algorithm is first used to preprocess the surface reflectance data of pure soil pixels. Then, the derivatives are calculated using the central difference method. The first derivative primarily enhances the clarity of the absorption valley outlines, while the second derivative further emphasizes the peak positions of the absorption valleys, making it particularly suitable for scenarios with low soil heavy metal content and weak spectral variation.

[0087] Continuum removal normalizes the surface reflectance data of pure soil pixels to the 0-1 interval by fitting the "envelope" of the spectral curve, eliminating differences in the overall spectral reflectance level and focusing on the relative intensity of spectral absorption features. The specific process is as follows: First, local maxima of the spectral curve are identified, and linear interpolation is used to connect these maxima to generate a continuum curve. Then, the original spectral reflectance is divided by the continuum reflectance at the corresponding wavelength to obtain the continuum-removed spectrum. After this transformation, the depth and shape of mineral absorption valleys related to heavy metals in the soil are significantly amplified, facilitating subsequent feature extraction.

[0088] Absorption depth analysis quantifies the characteristic parameters of absorption valleys in the removed spectrum of a continuum, transforming qualitative differences in spectral shape into quantitative characteristic indicators that directly correlate with the degree of influence of heavy metal content on soil spectra. Key parameters include absorption valley location, absorption depth, absorption width, and absorption area. Among these parameters, absorption depth and absorption area have the strongest correlation with heavy metal content (higher heavy metal content may lead to deeper and larger absorption valleys for related minerals), making them the focus of subsequent feature selection.

[0089] The three spectral transformation methods mentioned above can be used individually or in combination, and are suitable for different heavy metal types and soil textures.

[0090] The feature selection unit is used to use a genetic algorithm to screen out sensitive spectral features related to soil heavy metal content from spectrally enhanced surface reflectance data.

[0091] The feature selection unit is specifically used for:

[0092] The full-band set of spectrally enhanced surface reflectance data was used as the initial population.

[0093] The fitness value of an individual in the population is obtained by using the prediction accuracy of the heavy metal concentration inversion model constructed based on the selected band combination as the fitness function.

[0094] Based on fitness values, a roulette wheel selection method is used to select individuals for the next generation;

[0095] Perform crossover and mutation operations on individuals with preset probabilities to generate a new offspring population;

[0096] The selection, crossover, and mutation operations are performed iteratively until the preset number of generations is reached, and the combination of bands represented by the individuals with the highest fitness in each generation is used as the final selected sensitive spectral features.

[0097] In this embodiment, the spectral feature enhancement and extraction module successfully extracts a set of refined and strongly correlated sensitive spectral features from the original surface reflectance data of the test area through a three-step pipeline of "purifying the target → enhancing the signal → optimizing the screening". This lays a solid data foundation for the final realization of high-precision soil heavy metal pollution identification and dynamic early warning.

[0098] The system in this embodiment also includes:

[0099] The data analysis and inversion module is used to input sensitive spectral features into a preset intelligent inversion and early warning model library to obtain soil heavy metal pollution level identification results and quantitative concentration inversion results;

[0100] The data analysis and inversion module specifically includes:

[0101] The initial inversion unit is used to simultaneously input sensitive spectral features into the pollution level identification model and the quantitative concentration inversion model to obtain the initial pollution level and the initial quantitative concentration, respectively.

[0102] The pollution level identification model is a multi-classification model based on machine learning, and the quantitative concentration inversion model is a regression model based on machine learning.

[0103] The pollution level identification model and the quantitative concentration inversion model are jointly trained using a dual-task learning framework that shares a low-level feature network. The loss function during training is a weighted sum of regression loss and classification loss.

[0104] Specifically, the core design of the dual-task learning framework lies in "shared low-level feature extraction + independent task heads": the low-level feature network adopts 3-5 fully connected layers or a lightweight CNN, which is responsible for extracting general deep features from sensitive spectral features, such as the combination pattern of spectral absorption peaks and the correlation strength of feature bands. This part is trained by both types of tasks to avoid redundant calculations and enhance the generalization ability of features; the upper layers are set up with classification task heads (containing 2-3 fully connected layers + Softmax activation function) and regression task heads (containing 2-3 fully connected layers + linear activation function), respectively adapted to the probability output of grade recognition and the continuous value output of concentration inversion.

[0105] The loss function used in the training process is a weighted sum of regression loss and classification loss to achieve co-optimization of the two tasks: the classification loss uses cross-entropy loss to quantify the difference between the model's predicted level and the actual level; the regression loss uses root mean square error loss or mean absolute error loss to reflect the deviation between the predicted concentration and the measured concentration. The weight coefficients are dynamically adjusted according to the task priority. If the system focuses more on pollution level identification, the classification loss weight is set to 0.5-0.6, and the regression loss weight is set to 0.4-0.5; if the focus is on quantitative concentration inversion, the regression loss weight is increased to 0.5-0.6, and the classification loss weight is set to 0.4-0.5, ensuring that the loss function can guide the model to optimize the performance of both tasks simultaneously. The training data should adopt a "labeled sample + data augmentation" strategy: the labeled samples should include sensitive spectral features, corresponding measured heavy metal concentrations, and pollution level labels, and be divided into training set, validation set, and test set in a 7:2:1 ratio; data augmentation expands the training set by adding Gaussian noise (noise intensity ≤0.01), feature shifting (±1 band), etc., to avoid model overfitting. The training process is as follows: First, the shared underlying feature network is frozen, and the classification task head and regression task head are pre-trained separately until convergence. Then, the shared network is unfrozen, and end-to-end training is performed using a joint loss function. An early stopping strategy is used to prevent overfitting, ultimately resulting in a trained pollution level identification model and a quantitative concentration inversion model. During the initial inversion, sensitive spectral features are simultaneously input into both models. The classification model outputs the probability distribution of each pollution level, and the level with the highest probability is taken as the initial pollution level. The regression model directly outputs the specific heavy metal concentration value as the initial quantitative concentration.

[0106] The bidirectional feedback unit is used to map the initial quantitative concentration value to a predefined concentration range label, and to input the concentration range label as a new feature dimension, together with the sensitive spectral features, into the pollution level identification model for secondary discrimination, so as to output the final soil heavy metal pollution level identification result; it is also used to use the preset concentration threshold range corresponding to the initial pollution level as a constraint condition for the quantitative concentration inversion model, and to correct the initial quantitative concentration value that exceeds the range, so as to constrain it within the threshold range, so as to obtain the final quantitative concentration inversion result.

[0107] Specifically, the bidirectional feedback unit includes feedback flows in the following two directions:

[0108] The initial quantitative concentration values ​​themselves contain rich continuous scaling information. This unit first maps these continuous concentration values ​​to predefined pollution level interval labels according to national or industry standards; for example, a concentration value of 15 mg / kg corresponds to the "lightly polluted" label. Subsequently, this "concentration interval label," generated by a high-precision regression model, serves as a new and highly valuable information dimension, and is input again into the pollution level identification model for secondary discrimination along with the original sensitive spectral features. This process provides clear numerical clues for the classification model, effectively correcting potential category ambiguity or misjudgment that may occur when classifying solely based on spectral shape, thereby outputting a more accurate final soil heavy metal pollution level identification result.

[0109] Conversely, the initial pollution level results provide a reasonable concentration range constraint. Each pollution level corresponds to a theoretical concentration threshold range; for example, "moderate pollution" might correspond to a concentration range of [50, 100] mg / kg. This unit utilizes this prior knowledge to compare the initial quantitative concentration value with the preset threshold range of its corresponding initial pollution level. For "abnormal concentration values" that may be generated in the regression model due to extreme values ​​or noise and significantly exceed the reasonable range of their respective levels, the system will correct them to within the upper and lower limits of the threshold corresponding to that level. This operation is equivalent to adding a reasonableness check and smoothing constraint based on classification logic to the output of the regression model, ensuring that the final output quantitative concentration inversion result is not only numerically accurate but also logically consistent with the level, avoiding counterintuitive prediction results.

[0110] The aforementioned data analysis and inversion module, through its ingenious "joint training + two-way feedback" mechanism, successfully breaks down the traditional barriers between separate qualitative classification and quantitative regression models. It transforms qualitative and quantitative analysis from two parallel lines into a mutually reinforcing closed-loop optimization system, ultimately significantly improving the overall accuracy and practical value of soil heavy metal pollution identification and inversion, providing more reliable data support for precise environmental supervision and early warning.

[0111] The dynamic early warning module is used to analyze the spatiotemporal diffusion trend of soil heavy metal pollution and conduct dynamic risk assessment based on the soil heavy metal pollution level identification results and quantitative concentration inversion results, combined with multi-temporal satellite data and environmental auxiliary geographic data covering the area to be tested, and generate risk early warning reports.

[0112] The dynamic early warning module includes:

[0113] The spatiotemporal trend analysis unit is used to simulate the spatiotemporal diffusion trend of heavy metal pollution in soil based on the quantitative concentration inversion results of multiple time phases, using cellular automata models or geographically weighted regression models, and outputs a spatial distribution simulation map of pollutant concentration over a future period of time.

[0114] Among them, the cellular automata model is a dynamic simulation model based on "discrete spatiotemporal, local interaction, and state transition". Its core idea is to rasterize the research area into countless "cells", each cell has a clear state, such as heavy metal concentration level and pollution status. By defining neighborhood interaction rules and state transition rules, the state of each cell depends only on its own state at the previous moment and the state of neighboring cells. The spatiotemporal evolution process of the entire system is simulated through multiple rounds of iteration.

[0115] The geographically weighted regression model is a spatial extension of the traditional global regression model. Its core idea is to "acknowledge the spatial heterogeneity of geographical phenomena"—that is, the degree of influence of the independent variable on the dependent variable (regression coefficient) will change with spatial location. By assigning different weights to samples in different spatial locations, a local weighted regression equation is constructed, thereby accurately quantifying the local contribution of the independent variable in different regions and avoiding the error caused by the "one-size-fits-all" approach of the global regression model.

[0116] In the specific implementation process, the spatiotemporal trend analysis unit receives quantitative concentration inversion results covering multiple temporal phases of the area to be measured. That is, at different time points, such as different months, quarters or years, the quantitative concentration inversion results obtained by inverting hyperspectral data for the same area to be measured have temporal continuity and spatial consistency. Based on this, the spatiotemporal trend analysis unit uses cellular automata models or geographically weighted regression models to simulate and predict the spatial diffusion process of pollutants.

[0117] For example, when using a cellular automata model, the system discretizes the area to be measured into a regular grid, or cells, each cell having a state such as "pollution concentration." The model simulates the spatial diffusion, accumulation, or decay of pollutants by defining transformation rules. When using a geographically weighted regression model, by establishing a local regression equation for each pixel location, the spatial distribution pattern of pollution is simulated with high precision, revealing the underlying dominant driving mechanisms and their spatial variability, providing targeted information for targeted governance.

[0118] The dynamic risk assessment unit is used to integrate the pollution diffusion trend, pollution level identification results and environmental auxiliary geographic data to conduct pollutant migration simulation and risk assessment, and to delineate dynamic risk level areas.

[0119] The dynamic risk assessment unit includes:

[0120] The spatiotemporal sequence construction subunit is used to construct a pixel-level time series dataset of heavy metal concentration and pollution level based on the quantitative concentration inversion results and pollution level identification results of multiple time phases.

[0121] The core task of the spatiotemporal series construction subunit is to build a pixel-level, highly consistent time-series dataset of heavy metal concentrations and pollution levels, providing fundamental data support for subsequent early warning analysis. The data sources for this subunit include multi-temporal quantitative concentration inversion results, corresponding pollution level identification results, and synchronously collected environmental auxiliary geographic data, such as soil type, land use type, soil pH value, slope and aspect, pollution source distribution, and road network density.

[0122] The early warning analysis subunit is used to analyze pollution trends based on time series datasets using time series forecasting methods, and to perform diffusion simulations by combining environmental auxiliary geographic data, generating trend-based early warning and diffusion early warning information.

[0123] Furthermore, the early warning analysis subunit includes:

[0124] The trend warning analysis node is used to fit the time series dataset using an autoregressive integral moving average model or a long short-term memory network model, predict the trend of pollution concentration changes in a specific future period, and issue trend warning signals for areas that exceed the pollution risk level threshold.

[0125] The diffusion early warning analysis node is used to combine water flow direction and cumulative volume data derived from the digital elevation model, as well as regional prevailing wind direction frequency data, to simulate the spatial diffusion path and impact range of pollutants based on the heavy metal migration model, and generate diffusion early warning signals.

[0126] The trend early warning analysis node focuses on the time-dimensional trend of pollution development. It employs mature time-series prediction models, such as autoregressive integral moving average models or long short-term memory networks (LSMNs), which are better at capturing long-term dependencies and complex nonlinear patterns. The model fits the average concentration time-series data for each pixel or typical area to predict the pollution concentration level for a specific future period. By comparing this prediction with preset pollution risk thresholds at various levels, the system can issue trend-based early warning signals for areas that are about to exceed standards or where pollution levels are expected to continue to worsen. This early warning is based on direct extrapolation of historical patterns, answering the question, "What will happen in the future if the current trend remains unchanged?"

[0127] The diffusion early warning analysis node focuses on the spatial migration paths of pollution. It combines key environmental auxiliary geographic data to simulate the transport of pollutants in environmental media. For example, using water flow direction and accumulation data derived from digital elevation models, it can simulate the downstream migration paths and potential enrichment areas of heavy metals carried by surface water and soil particles under the influence of rainfall runoff. Simultaneously, by combining regional prevailing wind direction and frequency data, it can simulate the diffusion flux of pollutants downwind, primarily through atmospheric deposition or dust. By integrating these physical migration models, the node can predict the spatial diffusion paths of pollutants, their impact range, and potentially newly endangered sensitive targets, thereby generating diffusion early warning signals.

[0128] The dynamic early warning module also includes:

[0129] The report generation unit is used to generate a risk warning report based on the pollution diffusion trend and dynamic risk level area, including the spatial distribution of risk and key risk sources. This risk warning report is a dynamic risk warning report including risk maps, statistical charts, and text descriptions. This report not only clearly indicates the current level of risk but also warns of the risk's evolution trend in time and space, truly achieving a leap from "current situation description" to "trend prediction" and from "passive response" to "proactive prevention and control." It provides crucial scientific basis for land spatial planning, pollution source control, and the determination of priority remediation areas.

[0130] The aforementioned dynamic early warning module is the core output and decision support component of the intelligent identification and dynamic early warning system for soil heavy metal pollution. It receives pollution level identification and quantitative concentration inversion results from the data analysis and inversion module. Its core objective is to leverage the temporal continuity of multi-temporal satellite data and the spatial correlation of environmental auxiliary geographic data. Through a full-process analysis of "spatiotemporal trend modeling – risk level classification – precise early warning generation," it achieves full-chain tracing, simulation, and early warning of soil heavy metal pollution from the past to the present to the future. This provides dynamic, visualized, and actionable decision-making support for pollution prevention and control, source control, and risk avoidance. This module breaks through the limitations of traditional static assessments, emphasizing "spatiotemporal linkage" and "dynamic updates." It not only mines pollution evolution patterns through time-series data but also simulates diffusion paths based on geographical environmental characteristics, ultimately generating risk early warning reports that are both trend-based and forward-looking.

[0131] This application presents an intelligent identification and dynamic early warning system for heavy metal pollution in soil. By constructing a complete technology chain integrating data acquisition and preprocessing, spectral feature enhancement, intelligent inversion analysis, and dynamic early warning, it achieves a leap from high-precision identification to trend prediction of regional soil heavy metal pollution. Its core beneficial effects are reflected in the following: Based on multi-source satellite hyperspectral data and advanced fusion algorithms, it is the first to acquire surface reflectance data with both high spectral and spatial resolution, laying a data foundation for accurate inversion; through spectral purification, enhancement, and intelligent feature selection, it effectively extracts weak spectral signals related to heavy metal content; it innovatively adopts a dual-task learning and bidirectional feedback mechanism, enabling mutual verification and collaborative optimization of pollution level determination and quantitative concentration inversion results, significantly improving inversion accuracy and logical consistency; finally, by integrating multi-temporal data and environmental geographic elements, it utilizes a spatiotemporal dynamic model to realize pollution diffusion trend simulation and forward-looking risk early warning, pushing pollution monitoring from static assessment to dynamic management, providing scientific, efficient, and operable comprehensive technical support for soil pollution prevention and remediation decisions.

[0132] Example 2

[0133] Figure 2 This application describes a method for intelligent identification and dynamic early warning of heavy metal pollution in soil, according to one embodiment. Figure 2 As shown, the method includes:

[0134] Multi-source satellite hyperspectral data of the area to be measured are acquired, and the multi-source satellite hyperspectral data are preprocessed and fused to obtain the surface reflectance data of the area to be measured.

[0135] Soil pixel purification and spectral feature enhancement were performed on the surface reflectance data of the area to be tested to extract sensitive spectral features related to soil heavy metal content.

[0136] The sensitive spectral features are input into a preset intelligent inversion and early warning model library to obtain soil heavy metal pollution level identification results and quantitative concentration inversion results;

[0137] Based on the soil heavy metal pollution level identification results and quantitative concentration inversion results, combined with multi-temporal satellite data and environmental auxiliary geographic data covering the area to be tested, the spatiotemporal diffusion trend analysis and dynamic risk assessment of soil heavy metal pollution are carried out, and a risk warning report is generated.

[0138] The intelligent identification and dynamic early warning method for heavy metal pollution in soil proposed in this application obtains high-quality surface reflectance data through the fusion of multi-source satellite hyperspectral data. It accurately extracts sensitive spectral features of heavy metals through soil pixel purification and spectral feature enhancement, and uses an intelligent inversion model to simultaneously output pollution level and quantitative concentration results. Then, it combines multi-temporal data and environmental geographic elements to conduct spatiotemporal diffusion simulation and dynamic risk assessment, and finally generates a forward-looking risk warning report. This achieves closed-loop management from data acquisition to risk warning, significantly improving the accuracy of pollution identification, the scientific nature of trend prediction, and the timeliness of prevention and control decisions.

[0139] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0140] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0141] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "on top of," and "over" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0142] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0143] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A soil heavy metal pollution intelligent identification and dynamic early warning system, characterized in that, include: The data acquisition and preprocessing module is used to acquire multi-source satellite hyperspectral data of the area to be measured, and to perform data preprocessing and data fusion on the multi-source satellite hyperspectral data to obtain the surface reflectance data of the area to be measured. The spectral feature enhancement and extraction module is used to purify soil pixels and enhance spectral features of the surface reflectance data of the area to be tested, and extract sensitive spectral features related to soil heavy metal content. The data analysis and inversion module is used to input the sensitive spectral features into a preset intelligent inversion and early warning model library to obtain soil heavy metal pollution level identification results and quantitative concentration inversion results; The dynamic early warning module is used to analyze the spatiotemporal diffusion trend of soil heavy metal pollution and conduct dynamic risk assessment based on the soil heavy metal pollution level identification results and quantitative concentration inversion results, combined with multi-temporal satellite data and environmental auxiliary geographic data covering the area to be tested, and generate a risk early warning report.

2. The intelligent identification and dynamic early warning system for soil heavy metal pollution according to claim 1, characterized in that, The data acquisition and preprocessing module includes: A radiometric calibration unit is used to convert the multi-source satellite hyperspectral data into apparent radiance data; An atmospheric correction unit is used to process the apparent radiance data using a 6S or MODTRAN model to eliminate atmospheric effects and obtain initial surface reflectance data. The geometric correction unit is used to perform geometric fine correction on the initial surface reflectance data based on ground control points and digital elevation model, and obtain the geometrically corrected surface reflectance data. The data fusion unit is used to perform spatial-spectral fusion of the geometrically corrected surface reflectance data with the simultaneous multispectral data to obtain the surface reflectance data of the area to be measured.

3. The intelligent identification and dynamic early warning system for soil heavy metal pollution according to claim 1, characterized in that, The spectral feature enhancement and extraction module includes: The pure pixel extraction unit is used to classify the surface reflectance data of the area to be measured using vegetation index and water index, and remove pixels in vegetation-covered and water-covered areas by setting a threshold mask to extract the surface reflectance data of pure soil pixels. The spectral enhancement unit is used to perform at least one transformation, including spectral differentiation, continuum removal, or absorption depth analysis, on the surface reflectance data of the pure soil pixels to obtain spectrally enhanced surface reflectance data. The feature selection unit is used to use a genetic algorithm to screen out sensitive spectral features related to soil heavy metal content from spectrally enhanced surface reflectance data.

4. The intelligent identification and dynamic early warning system for soil heavy metal pollution according to claim 3, characterized in that, The feature selection unit is specifically used for: The full-band set of the spectrally enhanced surface reflectance data is used as the initial population. The fitness value of an individual in the population is obtained by using the prediction accuracy of the heavy metal concentration inversion model constructed based on the selected band combination as the fitness function. Based on the fitness value, an individual is selected to enter the next generation using roulette wheel selection. Perform crossover and mutation operations on the selected individuals with preset probabilities to generate a new offspring population; The selection, crossover, and mutation operations are performed iteratively until the preset number of generations is reached, and the combination of bands represented by the individuals with the highest fitness in each generation is used as the final selected sensitive spectral features.

5. The intelligent identification and dynamic early warning system for soil heavy metal pollution according to claim 1, characterized in that, The data analysis and inversion module includes: The initial inversion unit is used to simultaneously input the sensitive spectral features into the pollution level identification model and the quantitative concentration inversion model to obtain the initial pollution level and the initial quantitative concentration, respectively. The bidirectional feedback unit is used to map the initial quantitative concentration value to a predefined concentration range label, and to input the concentration range label as a new feature dimension, together with the sensitive spectral feature, into the pollution level identification model for secondary discrimination, so as to output the final soil heavy metal pollution level identification result; it is also used to use the preset concentration threshold range corresponding to the initial pollution level as the constraint condition of the quantitative concentration inversion model, and to correct the initial quantitative concentration value that exceeds the range, so as to constrain it within the threshold range, so as to obtain the final quantitative concentration inversion result.

6. The intelligent identification and dynamic early warning system for soil heavy metal pollution according to claim 5, characterized in that, The pollution level identification model is a multi-classification model based on machine learning, and the quantitative concentration inversion model is a regression model based on machine learning. The pollution level identification model and the quantitative concentration inversion model are jointly trained using a dual-task learning framework that shares a low-level feature network. The loss function during training is a weighted sum of regression loss and classification loss.

7. The intelligent identification and dynamic early warning system for soil heavy metal pollution according to claim 1, characterized in that, The dynamic early warning module includes: The spatiotemporal trend analysis unit is used to simulate the spatiotemporal diffusion trend of soil heavy metal pollution based on the quantitative concentration inversion results of multiple time phases, using a cellular automata model or a geographically weighted regression model. The dynamic risk assessment unit is used to integrate the pollution diffusion trend, pollution level identification results and environmental auxiliary geographic data to conduct pollutant migration simulation and risk assessment, and to delineate dynamic risk level areas. The report generation unit is used to generate a risk warning report containing the spatial distribution of risks and key risk sources based on the pollution diffusion trend and dynamic risk level area.

8. The intelligent identification and dynamic early warning system for soil heavy metal pollution according to claim 7, characterized in that, The dynamic risk assessment unit includes: The spatiotemporal sequence construction subunit is used to construct a pixel-level time series dataset of heavy metal concentration and pollution level based on the quantitative concentration inversion results and pollution level identification results of multiple time phases. The early warning analysis subunit is used to analyze pollution trends based on the time series dataset using time series forecasting, and to perform diffusion simulations in conjunction with environmental auxiliary geographic data, generating trend-based early warning and diffusion-based early warning information.

9. The intelligent identification and dynamic early warning system for soil heavy metal pollution according to claim 8, characterized in that, The early warning analysis subunit includes: The trend warning analysis node is used to fit the time series dataset using an autoregressive integral moving average model or a long short-term memory network model, predict the trend of pollution concentration changes in a specific future period, and issue trend warning signals for areas that exceed the pollution risk level threshold. The diffusion early warning analysis node is used to combine water flow direction and cumulative volume data derived from the digital elevation model, as well as regional prevailing wind direction frequency data, to simulate the spatial diffusion path and impact range of pollutants based on the heavy metal migration model, and generate diffusion early warning signals.

10. A method for intelligent identification and dynamic early warning of heavy metal pollution in soil, characterized in that, include: Multi-source satellite hyperspectral data of the area to be measured are acquired, and the multi-source satellite hyperspectral data are preprocessed and fused to obtain the surface reflectance data of the area to be measured. Soil pixel purification and spectral feature enhancement were performed on the surface reflectance data of the area to be tested to extract sensitive spectral features related to soil heavy metal content. The sensitive spectral features are input into a preset intelligent inversion and early warning model library to obtain soil heavy metal pollution level identification results and quantitative concentration inversion results; Based on the soil heavy metal pollution level identification results and quantitative concentration inversion results, combined with multi-temporal satellite data and environmental auxiliary geographic data covering the area to be tested, the spatiotemporal diffusion trend analysis and dynamic risk assessment of soil heavy metal pollution are carried out, and a risk warning report is generated.

Images