A mine area loess cadmium pollution identification method and system

Abstract

The application discloses a mine area loess cadmium pollution identification method, by acquiring the near-infrared spectrum information of montmorillonite after adsorbing different concentrations of cadmium ions, an identification benchmark directly representing the spectral response of the target mineral in loess caused by cadmium adsorption is established, the hyperspectral image is sequentially subjected to radiation correction and atmospheric correction, the resolution of the reflectivity image is improved, the accuracy of the geometric distribution of the surface object is further improved, the reflectivity image is further subjected to orthographic correction, the geometric distortion caused by the earth curvature, the surface topography and the satellite observation platform movement is eliminated, the geometric accuracy of the remote sensing data and the accuracy of the spatial positioning are further improved, according to the near-infrared spectrum information of montmorillonite after adsorbing different concentrations of cadmium ions combined with a deep learning algorithm, a cadmium pollution identification model is constructed, the defects of the traditional method, such as insufficient training samples and poor anti-interference in a complex mine area background, are overcome, and high-precision inversion of the spatial distribution of loess cadmium pollution is realized.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing identification technology for soil pollution, and relates to a method and system for identifying cadmium pollution in yellow clay soil in mining areas. Background Technology

[0002] Loess soil is formed gradually from loess parent material through long-term cultivation and utilization, and belongs to a relatively underdeveloped juvenile soil. It contains various clay minerals such as montmorillonite, chlorite, kaolinite, and vermiculite, with montmorillonite being particularly abundant, accounting for over 10%. In recent years, with the rapid socio-economic development and increasing intensity of mineral resource exploitation in the region, the natural ecological environment around many mining areas has been severely damaged, and the soil environment faces an increasingly serious threat of heavy metal pollution. Heavy metals are characterized by their difficulty in biodegradation, easy accumulation in soil, and long natural remediation cycles. Once they enter the environment, they persist for a long time and accumulate in the human body through biomagnification in the food chain, posing a serious and lasting risk to human health and safety. Currently, traditional soil sampling and laboratory analysis methods face challenges such as low efficiency, long cycles, and insufficient spatial coverage in rapidly identifying and accurately monitoring heavy metal pollution in soil under complex surface conditions, especially in mining areas, farmland, and ecologically fragile areas. These areas often have complex terrain and wide monitoring ranges. Relying solely on manual sampling is not only costly but also makes it difficult to achieve large-scale, continuous dynamic monitoring, resulting in the inability to obtain the spatiotemporal distribution characteristics of soil heavy metal pollution in a timely and accurate manner. Furthermore, in complex surface environments, factors such as vegetation cover, soil moisture changes, and surface background interference further increase the uncertainty of remote sensing monitoring. Traditional remote sensing data, limited by band settings and insufficient utilization of spectral information, struggles to effectively identify changes in heavy metal content in soil and the subtle differences between them and mineral composition.

[0003] Therefore, based on the above problems, there is an urgent need to provide a method for identifying cadmium pollution in loess soil in mining areas, which can combine near-infrared hyperspectral remote sensing and deep learning to determine the distribution of cadmium pollution in loess soil in mining areas, improve the rapid detection and identification capabilities of heavy metal pollution in soil, and provide theoretical basis and technical support for soil pollution monitoring, ecological environment governance and pollution remediation. Summary of the Invention

[0004] The purpose of this invention is to address the lack of a dedicated identification mechanism for the weak spectral signal response of montmorillonite target minerals in loess soil caused by cadmium adsorption in existing remote sensing identification methods. Furthermore, these methods are limited by the mismatch between satellite and ground spectral scales and insufficient training samples, making it difficult to realize the spatial distribution inversion of cadmium pollution in loess soil in complex mining areas. This invention provides a method and system for identifying cadmium pollution in loess soil in mining areas.

[0005] To achieve the above objectives, the present invention employs the following technical solution: A method for identifying cadmium pollution in loess soil in mining areas includes the following steps: Obtain near-infrared spectral information of montmorillonite after adsorption of different concentrations of cadmium ions; Obtain hyperspectral images of cadmium-contaminated areas in yellow clay soils; The hyperspectral image is then subjected to radiometric correction and atmospheric correction in sequence to obtain a reflectance image. Orthorectified hyperspectral image is obtained by orthorectifying the reflectance image; Based on near-infrared spectral information, a deep learning algorithm is introduced to construct a cadmium pollution identification model. Based on the cadmium pollution identification model, cadmium pollution in orthorectified hyperspectral images is identified, and the identification results are obtained.

[0006] A further improvement of the present invention is that: The acquisition of near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions includes: Near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions was collected using a near-infrared spectrometer. The acquired near-infrared spectral information is preprocessed, and nonlinear least squares fitting is performed on the preprocessed near-infrared spectral data to perform peak decomposition and spectrum resolution, thus obtaining the processed near-infrared spectral information.

[0007] The process of sequentially performing radiometric and atmospheric corrections on the hyperspectral image to obtain a reflectance image includes the following steps: Radiation correction is performed using the following formula:

[0008] in, This is the radiance value. For gain, For bias, This refers to the pixel brightness value. Atmospheric correction is performed using the following formula:

[0009] in, The total radiance received by the sensor pixel. The surface reflectivity of a pixel. The average surface reflectance around the pixel. The albedo of the balloon surface. This represents the atmospheric backscattered emissivity. This is the correlation coefficient for atmospheric conditions. This is the geometric conditional correlation coefficient.

[0010] The process of orthorectifying the reflectance image to obtain an orthorectified hyperspectral image includes the following steps: Orthorectification is performed using the following formula:

[0011] Where (X,Y,Z) are the geographic coordinates of the ground point (elevation Z needs to be provided by the DEM), , , (x, y) represents the satellite position, R is the attitude rotation matrix, roll is the roll angle, pitch is the pitch angle, yaw is the yaw angle, and (x, y) are the image point coordinates. It is the focal length.

[0012] The cadmium pollution identification model constructed based on near-infrared spectral information and incorporating deep learning algorithms includes: The raw near-infrared spectral data is converted to units and matched to dimensions to unify the wavenumbers into wavelength units. By combining the spectral response parameters of the remote sensing sensor, several target feature points are set in the 2000–2400 nm shortwave infrared region. The continuous spectral lines are resampled using a linear interpolation algorithm to obtain the reflectance value at the target wavelength. Based on the relationship curve between adsorption amount and spectral shift, near-infrared spectral data with random noise under different adsorption conditions are generated by combining reflectance values. A cadmium pollution identification model was constructed by training a near-infrared spectral label dataset and corresponding cadmium ion concentrations using the HybridSN deep learning algorithm.

[0013] The method of using the HybridSN deep learning algorithm to train a cadmium pollution identification model on a near-infrared spectral label dataset and corresponding cadmium ion concentrations includes: A dual-branch parallel convolutional structure is used in the input layer to extend the traditional single hyperspectral cubic input to a dual-stream input mode of proprietary spectral data and near-infrared spectrum. Spatial spectral features are extracted using a three-layer 3D convolution. A spectral peak attention mechanism is introduced after the 3D convolution layer. The input patch size is 15×15 and contains 48 feature bands.

[0014] A system for identifying cadmium pollution in yellow clay soil in mining areas includes: The near-infrared spectroscopy acquisition module is used to acquire near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions; The hyperspectral image acquisition module is used to acquire hyperspectral images of cadmium-contaminated areas in yellow clay soil. The radiation and atmospheric correction module is used to perform radiometric and atmospheric corrections on the hyperspectral image sequentially to obtain a reflectance image. The orthorectification module is used to orthorectify reflectance images to obtain orthorectified hyperspectral images. The identification module is used to construct a cadmium pollution identification model based on near-infrared spectral information and deep learning algorithms. Based on the cadmium pollution identification model, the cadmium pollution in orthorectified hyperspectral images is identified to obtain the identification results.

[0015] A computer program product includes a computer program that, when executed by a processor, implements any one of the methods described.

[0016] A terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of any of the methods described above.

[0017] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the methods described herein.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a method for identifying cadmium pollution in loess soil in mining areas. By acquiring near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions, an identification benchmark is established to directly characterize the spectral response of target minerals in loess soil caused by cadmium adsorption. Radiometric and atmospheric corrections are sequentially applied to the hyperspectral images, improving the resolution of the reflectance images and further enhancing the accuracy of the geometric distribution of surface objects. Orthorectification of the reflectance images eliminates geometric distortions caused by Earth curvature, surface topography, and satellite observation platform movement, further improving the geometric accuracy and spatial positioning accuracy of remote sensing data. Based on the near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions, combined with a deep learning algorithm, a cadmium pollution identification model is constructed. This overcomes the shortcomings of traditional methods, such as insufficient training samples and poor anti-interference capabilities in complex mining areas, achieving high-precision inversion of the spatial distribution of cadmium pollution in loess soil, enabling rapid identification and monitoring of cadmium pollution in mining areas.

[0019] Furthermore, in this invention, when constructing the cadmium pollution identification model, the dual-stream input mode effectively solves the feature loss problem in data fusion, enhancing the model's feature discrimination ability under complex mineral compositions. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of a method for identifying cadmium pollution in loess soil in a mining area, as described in this embodiment. Figure 2 This is a flowchart of the near-infrared spectral data acquisition process after montmorillonite adsorbs different concentrations of cadmium ions in this embodiment; Figure 3 This is a flowchart of the near-infrared spectral data preprocessing process after montmorillonite adsorbs different concentrations of cadmium ions in this embodiment. Figure 4a This is a near-infrared spectrum preprocessing result of samples 1 to 3 in this embodiment after montmorillonite adsorbed different concentrations of cadmium ions; Figure 4b This is a near-infrared spectrum preprocessing result of samples 4 to 6 in this embodiment after montmorillonite adsorbed different concentrations of cadmium ions; Figure 4c This is a near-infrared spectrum preprocessing result of samples 7 to 9 in this embodiment after montmorillonite adsorbed different concentrations of cadmium ions; Figure 5a This is a flowchart of the hyperspectral image processing of the Gaofen-5 domestic satellite in this embodiment; Figure 5b This is a flowchart of the orthorectified hyperspectral image processing in this embodiment; Figure 6 This is a diagram showing the identification results of cadmium pollution in the yellow clay soil of the experimental area in this embodiment; Figure 7 This is a Kriging interpolation map showing the cadmium pollution identification results of the yellow loess soil in the experimental area in this embodiment; Figure 8 This is a Kriging interpolation result of cadmium pollution in actual loess soil samples from the experimental area in this embodiment. Figure 9 This is a spatial distribution image of the difference between the identification result and the measured value in this embodiment. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0024] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0025] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0027] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; 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; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0028] The present invention will now be described in further detail with reference to the accompanying drawings: See Figures 1 to 9 This invention discloses a method for identifying cadmium pollution in loess soil in mining areas. It can combine near-infrared hyperspectral remote sensing and deep learning to determine the distribution of cadmium pollution in loess soil in mining areas, improve the rapid detection and identification capabilities of heavy metal pollution in soil, and provide theoretical basis and technical support for soil pollution monitoring, ecological environment governance and pollution remediation.

[0029] Specifically, the following steps are included: Step 1: Obtain near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions.

[0030] This step specifically includes the following processes: Step 11: Use a Lambda 950 near-infrared spectrometer to collect near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions.

[0031] Currently, the main experimental methods for near-infrared spectroscopy are reflection and transmission methods: The reflection method involves shining a near-infrared light source onto the sample surface and then measuring the reflected light signal. The transmission method involves passing the light source through the sample and measuring the light transmitted to the other side. The former is more suitable for solid samples such as powders and granules, while the latter is more suitable for transparent materials such as liquids and thin films.

[0032] In this embodiment, the montmorillonite used in the experiment is an opaque mineral material, so near-infrared spectral reflectance method was selected for detection.

[0033] The instrument used in this embodiment is a Lambda 950 near-infrared spectrometer manufactured by PerkinElmer, Inc. The light source of the instrument is provided by a halogen lamp. During the test, a white reflector made of polytetrafluoroethylene is used to convert the raw data into relative reflectance. The instrument has a spectral resolution of 4 nm, a wavelength accuracy of ±0.08 nm, a repeatability error of less than 0.005 nm, and stray light of no more than 0.00007%.

[0034] Step 12: Preprocessing of the acquired near-infrared spectral data; the preprocessing includes: signal averaging, spectral smoothing, feature clipping, dimension unification, and unit conversion.

[0035] This embodiment preprocesses the acquired near-infrared spectral data by first performing at least 15 consecutive repeated scans of the sample and then arithmetically averaging the resulting raw waveforms to reduce random errors caused by electronic noise and obtain a high signal-to-noise ratio average spectrum that characterizes the inherent properties of the sample. Furthermore, to address residual spikes caused by factors such as airflow fluctuations, instrument baseline drift, and transient instability of the light source in the experimental environment, a smoothing and denoising algorithm is introduced to further optimize the spectral lines, filtering out high-frequency clutter and enhancing the stability of the spectral curves, thereby improving the accuracy of identifying weak characteristic peaks.

[0036] Furthermore, to reduce redundant information interference and enhance the specificity of the analysis, the full-band spectrum was cropped to 4400–5000 cm⁻¹ based on the sensitive range of montmorillonite structural response. -1 The range is used for subsequent quantitative analysis. Meanwhile, to achieve consistency with traditional molecular spectroscopy and mid-infrared fundamental frequency data in spectral band assignment, the original wavelength units (nm) are uniformly converted to wavenumber units (cm). -1 Furthermore, based on A = 1 The linear mapping relationship of R further converts the collected reflectance data into an absorptivity mode. This process only involves redefining the spectral intensity dimension and does not change the frequency position of the absorption peak, thus ensuring the physical consistency of subsequent feature extraction and mechanism analysis.

[0037] Step 13: Use the Origin platform to perform nonlinear least squares fitting on the preprocessed near-infrared spectral data to perform peak separation and spectrum decomposition.

[0038] By using data analysis on the Origin platform, nonlinear least-squares fitting was performed on the overlapping peak clusters, and the composite envelope peaks were decomposed into physically meaningful independent sub-peaks. This was done to isolate characteristic components affected by cadmium ion loading, thereby quantitatively evaluating the response of the montmorillonite lattice to heavy metal ions.

[0039] Step 2: Obtain hyperspectral image data of the cadmium-contaminated area of ​​yellow cotton soil using the Gaofen-5 (GF-5) domestic satellite.

[0040] High-resolution (GF-5) hyperspectral imagery of the mining area, classified as Level L1 (ID: 0000076580), was downloaded from the Natural Resources Satellite Remote Sensing Cloud Service Platform. The GF-5 platform integrates six advanced detection payloads, including an Adaptive High-Speed ​​Imager (AHSI), a Full-Spectrum Imager (VIMS), and an Atmospheric Greenhouse Gas Monitor (GMI). For the soil heavy metal identification task conducted in this embodiment, high-quality data acquired using the AHSI payload was studied. The spectral range (nm) is 400-2500, with 318 channels, a spatial resolution of 30 m, and spectral resolutions of VNIR: 5 nm and SWIR: 10 nm.

[0041] Step 3: Perform radiometric correction and atmospheric correction on the hyperspectral image in sequence to obtain the reflectance image.

[0042] The digital orthophoto image is subjected to radiometric correction and atmospheric correction in sequence. Radiometric correction is to calibrate the original pixel values ​​of the hyperspectral image to radiometric values. The radiometrically corrected image needs to be subjected to atmospheric correction to obtain a reflectance image with pixel values ​​as reflectance.

[0043] The formula for radiation correction is as follows: ; in, This is the radiance value. For gain, For bias, This represents the pixel brightness value.

[0044] Furthermore, the atmospheric correction calculation formula is as follows: ; in, The total radiance received by the sensor pixel. The surface reflectivity of a pixel. The average surface reflectance around the pixel. The albedo of the balloon surface. This represents the atmospheric backscattered emissivity. This is the correlation coefficient for atmospheric conditions. This is the geometric conditional correlation coefficient.

[0045] Step 4: Perform orthorectification on the reflectance image to obtain an orthorectified hyperspectral image.

[0046] Orthorectification is the process of resampling pixels in remote sensing images into the correct geographic coordinate system using formulas and digital elevation models (DEMs) to eliminate geometric distortions caused by the curvature of the Earth, surface topography, and the motion of satellite observation platforms.

[0047] The formula for calculating orthorectification is as follows:

[0048] Where (X,Y,Z) are the geographic coordinates of the ground point (elevation Z needs to be provided by the DEM), , , ) represents the satellite position, R represents the attitude rotation matrix, roll represents the roll angle, pitch represents the pitch angle, and yaw represents the yaw angle. x , y () represents the coordinates of the image point. It is the focal length.

[0049] Step 5: Based on the near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions, a cadmium pollution identification model is constructed using the HybridSN deep learning algorithm.

[0050] Step 5 specifically includes: Step 51: Using a linear interpolation spectral resampling script, create the near-infrared spectral label dataset required for modeling; Step 51 specifically includes: Step 511: A spectral resampling method based on linear interpolation is used to perform unit conversion and dimension matching on the original near-infrared spectral data, and the wavenumber is uniformly converted into wavelength units; Step 512: Combining the spectral response parameters of the remote sensing sensor, 48 target feature points are set in the key region of 2000–2400 nm shortwave infrared (SWIR). The continuous spectral lines are resampled by a linear interpolation algorithm to obtain the accurate reflectance value at the target wavelength. Step 513: Based on the relationship curve between adsorption amount and spectral shift, generate 300 simulated near-infrared spectral data with random noise under different adsorption conditions using the uniform interpolation method.

[0051] Step 52: Use the HybridSN deep learning algorithm to train the near-infrared spectral label dataset and the corresponding cadmium ion concentration to build a cadmium pollution identification model; Step 52 specifically includes: Step 521: Use a dual-branch parallel convolutional structure in the input layer to extend the traditional single hyperspectral cube input into a dual-stream input mode of proprietary spectral data and near-infrared spectroscopy (NIR). Step 522: Spatial-spectral features are extracted using three layers of 3D convolution (with kernel sizes of 7×3×3, 5×3×3 and 3×3×3 respectively). A spectral peak attention mechanism is introduced after the 3D convolutional layer. The input patch size is 15×15 and contains 48 feature bands.

[0052] Step 523: The loss function used is the cross-entropy loss function, the optimizer is the Adam optimizer, and the initial learning rate is set to 5×10. 4. A weight decay coefficient of 1×10 is introduced. 4. To suppress overfitting, the dataset is divided into a training set (80%) and a validation set (20%) using stratified sampling. The batch size is set to 32, the epoch is set to 600, and early stopping is enabled.

[0053] Step 6: Use the cadmium pollution identification model to identify the distribution of cadmium pollution in the current orthorectified hyperspectral image.

[0054] The remote sensing identification method disclosed in this embodiment can be applied to the identification of cadmium pollution distribution in heavy metal contaminated mining areas where loess soil is widely distributed. Compared with traditional sampling and detection methods, this method can reduce the manpower, material resources, and time costs in large-scale surveys, while overcoming the problems of spatially dispersed sampling points and difficulty in achieving continuous regional monitoring, thus improving the efficiency and spatial coverage of cadmium pollution identification in mining areas. This embodiment reduces the cost of identifying cadmium pollution in loess soil in mining areas. This embodiment relies on the natural resources satellite remote sensing cloud service platform to acquire hyperspectral image data and quickly perform radiometric correction, atmospheric correction, and orthorectification. Combined with near-infrared spectral data after montmorillonite adsorption of different concentrations of cadmium ions and the HybridSN deep learning algorithm, it can efficiently and accurately identify the spatial distribution of cadmium pollution in heavy metal contaminated areas of loess soil in mining areas in a short time. In practical engineering, it can further identify the influencing factors of soil heavy metal pollution and provide remediation suggestions to improve the soil environmental quality of mining areas.

[0055] Furthermore, in an exemplary embodiment, a mining area with widespread loess soil and severe heavy metal pollution in a certain plain was selected as the experimental area. The total area of ​​the plain is 526 square kilometers, and 86.79% of the arable land is loess soil. The terrain generally shows a step-like descent from south to north.

[0056] Using a Lambda 950 near-infrared spectrometer according to Figure 2 The process involved collecting near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions. The collected near-infrared spectral data were then processed according to... Figure 3 The process involved signal averaging, spectral smoothing, feature trimming, dimensional unification, and unit conversion. The cadmium ion adsorption capacity of montmorillonite is shown in Table 1, and the corresponding near-infrared spectral preprocessing results are shown in [Table 1]. Figures 4a to 4c .

[0057] Table 1. Cd ion adsorption capacity

[0058] Step 1: Download the Gaofen-5 domestic satellite hyperspectral image of the mining area with a data product level of L1 (ID: 0000076580) using the Natural Resources Satellite Remote Sensing Cloud Service Platform. The hyperspectral image is shown in part 5a of Figure 5. In part 5a of Figure 5, the horizontal direction is longitude (east longitude) and the vertical direction is latitude (north latitude). Step 2: Perform radiometric and atmospheric corrections on the hyperspectral image to obtain a reflectance image. Perform orthorectification on the reflectance image to obtain an orthorectified hyperspectral image, as shown in part 5b of Figure 5. Step 3: Based on the linear interpolation-based spectral resampling method, the original near-infrared spectral data is processed by unit conversion and dimension matching to convert the wavenumber to wavelength units. Combined with the spectral response parameters of the remote sensing sensor, 48 target feature points are set in the key region of 2000–2400 nm short-wave infrared (SWIR). The continuous spectral lines are resampled by the linear interpolation algorithm to obtain the accurate reflectance value at the target wavelength. Based on the relationship curve between adsorption amount and spectral shift, 600 simulated near-infrared spectral data with random noise under different adsorption conditions are generated using the uniform interpolation method. The HybridSN deep learning algorithm is used to train the near-infrared spectral label dataset and the corresponding cadmium ion concentration to build a cadmium pollution identification model. Step 4: Construct a confusion matrix for the model training set (80%) and validation set (20%), using whether the model falls into the target category as the binary classification criterion, to evaluate the model results. The calculation results are shown in Table 2. The results show that the model's precision, recall, and F1 score are 90.68%, 87.60%, and 89.11%, respectively, with an overall accuracy of 88.29% and a Kappa coefficient of 0.765, indicating that it has good classification performance under the current sample conditions.

[0059] Table 2. Results of confusion matrix calculation

[0060] Step 5: Use the cadmium pollution identification model to identify the distribution of cadmium pollution in the current orthorectified hyperspectral image. The identification results are shown in [link to relevant documentation]. Figure 6 .

[0061] Furthermore, the spatial distribution of the inversion results in the study area was displayed using the Kriging interpolation method, as shown in the following figures. Figure 7 The measured cadmium concentrations in the soil at the designated sampling points were verified in the field. The Kriging interpolation results of the cadmium concentrations in the soil at the sampling points are shown below. Figure 8 The two Kriging interpolation results were converted into raster data of uniform resolution. By normalizing the data and performing raster subtraction, the spatial distribution image of the difference between the recognition result and the measured value was obtained, as shown below. Figure 9 As shown in Table 3, for the difference image, ArcMap raster calculation was used, and reclassification technology was employed to divide the error into five levels according to intervals. The pixel area occupied by each error interval was statistically analyzed. The specific statistical results are shown in Table 3. Data analysis shows that the area of ​​the region with a difference within 0.15 accounts for 48%, and the cumulative area of ​​the region with a difference within 0.25 accounts for as high as 81%. This indicates that the model has high recognition accuracy in more than 80% of the entire region.

[0062] Table 3. Statistics of pixel area corresponding to the difference

[0063] This embodiment uses a Lambda 950 near-infrared spectrometer to acquire near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions. Hyperspectral image data of the cadmium-contaminated area in loess soil is acquired from the Gaofen-5 domestic satellite. Simultaneously, radiometric and atmospheric corrections are performed on the hyperspectral images to obtain reflectance images. The correction process improves the resolution of the reflectance images, further enhancing the accuracy of the geometric distribution of surface objects. Orthorectification of the reflectance images is then performed using formulas and a digital elevation model (DEM) to obtain orthorectified hyperspectral images. This correction process eliminates geometric distortions caused by Earth curvature, surface topography, and satellite observation platform motion, further improving the geometric accuracy and spatial positioning accuracy of the remote sensing data. Based on the near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions, a cadmium pollution identification model was constructed using the HybridSN deep learning algorithm. Simulated near-infrared spectral data effectively overcomes the problem of insufficient dataset quantity in small-sample training. The HybridSN deep learning network effectively overcomes the limitations of traditional models in handling complex background noise. The dual-stream input mode effectively solves the feature loss problem in data fusion and enhances the model's feature discrimination ability under complex mineral compositions. Using the cadmium pollution identification model to identify the distribution of cadmium pollution in current orthorectified hyperspectral images enables rapid identification and monitoring of cadmium pollution in mining areas, improving source control of cadmium pollution and enhancing the soil environmental quality of mining areas.

[0064] This embodiment discloses a cadmium pollution identification system for loess soil in mining areas, including: The near-infrared spectroscopy acquisition module is used to acquire near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions; The hyperspectral image acquisition module is used to acquire hyperspectral images of cadmium-contaminated areas in yellow clay soil. The radiation and atmospheric correction module is used to perform radiometric and atmospheric corrections on the hyperspectral image sequentially to obtain a reflectance image. The orthorectification module is used to orthorectify reflectance images to obtain orthorectified hyperspectral images. The identification module is used to construct a cadmium pollution identification model based on near-infrared spectral information and deep learning algorithms. Based on the cadmium pollution identification model, the cadmium pollution in orthorectified hyperspectral images is identified to obtain the identification results.

[0065] A schematic diagram of a terminal device according to an embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the various method embodiments described above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0066] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention.

[0067] The terminal device can be a desktop computer, laptop computer, cloud server, or other device with strong computing power. The terminal device may include, but is not limited to, a processor and memory.

[0068] The optimal choice for the processor is a multi-core high-speed central processing unit (CPU).

[0069] The memory can be used to store the computer program and / or module. The processor implements various functions of the terminal device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0070] If the modules / units integrated into the terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0071] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying cadmium pollution in loess soil in mining areas, characterized in that, Includes the following steps: Obtain near-infrared spectral information of montmorillonite after adsorption of different concentrations of cadmium ions; Obtain hyperspectral images of cadmium-contaminated areas in yellow clay soils; The hyperspectral image is then subjected to radiometric correction and atmospheric correction in sequence to obtain a reflectance image. Orthorectified hyperspectral image is obtained by orthorectifying the reflectance image; Based on near-infrared spectral information, a deep learning algorithm is introduced to construct a cadmium pollution identification model. Based on the cadmium pollution identification model, cadmium pollution in orthorectified hyperspectral images is identified, and the identification results are obtained.

2. The method for identifying cadmium pollution in loess soil in mining areas according to claim 1, characterized in that, The acquisition of near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions includes: Near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions was collected using a near-infrared spectrometer. The acquired near-infrared spectral information is preprocessed, and nonlinear least squares fitting is performed on the preprocessed near-infrared spectral data to perform peak decomposition and spectrum resolution, thus obtaining the processed near-infrared spectral information.

3. The method for identifying cadmium pollution in loess soil in mining areas according to claim 1, characterized in that, The process of sequentially performing radiometric and atmospheric corrections on the hyperspectral image to obtain a reflectance image includes the following steps: Radiation correction is performed using the following formula: in, This is the radiance value. For gain, For bias, This refers to the pixel brightness value. Atmospheric correction is performed using the following formula: in, The total radiance received by the sensor pixel. The surface reflectivity of a pixel. The average surface reflectance around the pixel. The albedo of the balloon surface. This represents the atmospheric backscattered emissivity. This is the correlation coefficient for atmospheric conditions. This is the geometric conditional correlation coefficient.

4. The method for identifying cadmium pollution in loess soil in mining areas according to claim 1, characterized in that, The process of orthorectifying the reflectance image to obtain an orthorectified hyperspectral image includes the following steps: Orthorectification is performed using the following formula: Where (X,Y,Z) are the geographic coordinates of the ground point (elevation Z needs to be provided by the DEM), , , (x,y) represents the satellite position, R represents the attitude rotation matrix, roll represents the roll angle, pitch represents the pitch angle, yaw represents the yaw angle, and (x,y) represents the image point coordinates. It is the focal length.

5. The method for identifying cadmium pollution in loess soil in mining areas according to claim 1, characterized in that, The cadmium pollution identification model constructed based on near-infrared spectral information and incorporating deep learning algorithms includes: The raw near-infrared spectral data is converted to units and matched to dimensions to unify the wavenumbers into wavelength units. By combining the spectral response parameters of the remote sensing sensor, several target feature points are set in the 2000–2400 nm shortwave infrared region. The continuous spectral lines are resampled using a linear interpolation algorithm to obtain the reflectance value at the target wavelength. Based on the relationship curve between adsorption amount and spectral shift, near-infrared spectral data with random noise under different adsorption conditions are generated by combining reflectance values. A cadmium pollution identification model was constructed by training a near-infrared spectral label dataset and corresponding cadmium ion concentrations using the HybridSN deep learning algorithm.

6. The method for identifying cadmium pollution in loess soil in mining areas according to claim 5, characterized in that, The method of using the HybridSN deep learning algorithm to train a cadmium pollution identification model on a near-infrared spectral label dataset and corresponding cadmium ion concentrations includes: A dual-branch parallel convolutional structure is used in the input layer to extend the traditional single hyperspectral cubic input to a dual-stream input mode of proprietary spectral data and near-infrared spectrum. Spatial spectral features are extracted using a three-layer 3D convolution. A spectral peak attention mechanism is introduced after the 3D convolution layer. The input patch size is 15×15 and contains 48 feature bands.

7. A system for identifying cadmium pollution in loess soil in mining areas, characterized in that, include: The near-infrared spectroscopy acquisition module is used to acquire near-infrared spectral information of montmorillonite after adsorbing different concentrations of cadmium ions; The hyperspectral image acquisition module is used to acquire hyperspectral images of cadmium-contaminated areas in yellow clay soil. The radiation and atmospheric correction module is used to perform radiometric and atmospheric corrections on the hyperspectral image sequentially to obtain a reflectance image. The orthorectification module is used to orthorectify reflectance images to obtain orthorectified hyperspectral images. The identification module is used to construct a cadmium pollution identification model based on near-infrared spectral information and deep learning algorithms. Based on the cadmium pollution identification model, the cadmium pollution in orthorectified hyperspectral images is identified to obtain the identification results.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-6.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

Images