Multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation system and method

Abstract

The invention belongs to the cross technical field of remote sensing data processing and soil environment monitoring, and discloses a multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation system and method, and the method comprises the following steps: obtaining different types of remote sensing data in a target area, and constructing a multi-source data set; the acquired multi-source remote sensing data are preprocessed, data noise and system errors are eliminated, and data consistency is ensured; performing multi-source remote sensing data fusion by adopting a spectrum-polarization-texture collaborative fusion strategy in combination with a multi-scale decomposition and weighted fusion model; based on the fusion data, constructing a machine learning inversion model, and inverting soil quality key indexes; and carrying out comprehensive evaluation on the soil quality by adopting an analytic hierarchy process-matter element extension model to realize grading of the soil quality. By adopting the multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation system and method disclosed by the invention, rapid, accurate, large-area and dynamic monitoring and comprehensive evaluation of soil quality are realized.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of remote sensing data processing and soil environmental monitoring, and in particular to a system and method for inverting and comprehensively evaluating soil quality indicators using multi-source remote sensing collaboration. Background Technology

[0002] Soil quality is the core foundation for sustainable agricultural development and ecological environment stability. Its evaluation results provide key decision-making basis for land use planning, farmland protection, and pollution control. Traditional soil quality assessment mainly relies on manual field sampling and laboratory chemical analysis, which requires the measurement of soil organic matter, nitrogen, phosphorus, and potassium nutrients, pH value, heavy metal content, and other indicators. This method has the following limitations: First, the number of sampling points is limited, making it difficult to reflect the spatial heterogeneity of regional soil quality and limiting the coverage area; second, it is time-consuming and labor-intensive, with cumbersome sampling, transportation, and testing procedures that take several weeks, failing to meet the needs of real-time monitoring; third, it is costly, with high costs for testing equipment and reagents, making large-scale evaluation uneconomical; and fourth, it involves destructive sampling, which may disturb the soil ecosystem.

[0003] With the development of remote sensing technology, its advantages of macroscopic, rapid, and non-contact monitoring are increasingly being applied to soil quality assessment. Currently, commonly used remote sensing data sources include multispectral remote sensing (such as MODIS, Landsat, Sentinel-2), hyperspectral remote sensing (such as Hyperion, HJ-1A / 1B), and synthetic aperture radar (SAR, such as Sentinel-1, ALOS). However, single remote sensing data sources have inherent limitations: multispectral data has low spectral resolution, making it difficult to distinguish components with similar spectral characteristics in the soil; while hyperspectral data is rich in spectral information, it is susceptible to atmospheric noise and topographical influences, and its large data volume and high computational complexity are also significant; SAR data is sensitive to soil moisture and roughness, but has weak capabilities for inverting soil chemical properties. A single data source cannot comprehensively capture multidimensional information related to soil quality, such as spectral, structural, and textural aspects, leading to insufficient accuracy in soil quality index inversion and limited reliability of assessment results.

[0004] Therefore, how to construct an accurate soil quality assessment model by synergistically integrating multi-source remote sensing data and complementing the advantages of each data source has become a key technical problem in overcoming the limitations of traditional methods and single remote sensing assessment. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation system and method to solve the problems of small coverage, long time consumption, high cost and insufficient evaluation accuracy of single remote sensing data source in traditional soil quality evaluation, so as to realize rapid, accurate, large-area, dynamic monitoring and comprehensive evaluation of soil quality.

[0006] To achieve the above objectives, this invention provides a method for inverting and comprehensively evaluating soil quality indicators using multi-source remote sensing, comprising the following steps: Step S1: Acquire different types of remote sensing data within the target area and construct a multi-source dataset; Step S2: Preprocess the acquired multi-source remote sensing data to eliminate data noise and system errors, and ensure data consistency; Step S3: Employ a spectral-polarization-texture collaborative fusion strategy, combined with a multi-scale decomposition and weighted fusion model, to fuse multi-source remote sensing data; Step S4: Based on the fused data, construct a machine learning inversion model to invert key soil quality indicators; Step S5: Use the analytic hierarchy process (AHP)-matter extension model to conduct a comprehensive evaluation of soil quality and classify soil quality grades.

[0007] Preferably, in step S1, different types of remote sensing data within the target area are acquired to construct a multi-source data set, specifically including: multispectral remote sensing data, hyperspectral remote sensing data, SAR remote sensing data, and auxiliary data; The auxiliary data includes: digital elevation models of the target area, land use type maps, and meteorological data.

[0008] Preferably, in step S2, the acquired multi-source remote sensing data is preprocessed to eliminate data noise and systematic errors, ensuring data consistency. The specific process is as follows: Step S21: Radiometric calibration converts the grayscale values ​​of the remote sensing image into apparent radiance or reflectance, eliminating differences in the sensor's own response. Step S22: Use the MODTRAN atmospheric radiative transfer model to perform atmospheric correction, eliminate the influence of atmospheric scattering and absorption on remote sensing images, and obtain the true surface reflectance. Step S23: Use the polynomial correction method to perform geometric correction to eliminate image distortion caused by terrain undulation and sensor attitude error; Establish image pixel coordinates with geographic coordinates The polynomial mapping relationship is used to solve the coefficients using the least squares method, and the corrected image is resampled using bilinear interpolation. Step S24: Through data resampling and registration, unify the multi-source remote sensing data to the same spatial resolution and geographic coordinate system; The nearest neighbor method is used to upsample low-resolution data, and the bilinear interpolation method is used to downsample high-resolution data. Based on the mutual information registration algorithm, SAR data and hyperspectral data are registered with Sentinel-2 imagery as the reference, and the registration error is controlled within 0.5 pixels. Step S25: Use the Lee filtering algorithm to denoise and polarize the SAR data to remove speckle noise from the SAR data; The Yamaguchi polarization decomposition method was used to extract SAR polarization characteristic parameters, specifically including volume scattering power. Surface scattering power and secondary scattering power .

[0009] Preferably, in step S3, a spectral-polarization-texture collaborative fusion strategy is adopted, combined with a multi-scale decomposition and weighted fusion model, to fuse multi-source remote sensing data, thereby achieving complementary advantages of multi-source remote sensing data. The specific process is as follows: Step S31: Feature extraction; (1) Spectral features: Spectral features are extracted from multispectral / hyperspectral data, including vegetation index, soil index and hyperspectral absorption depth; (2) Texture features: Based on the gray-level co-occurrence matrix (GLCM), texture features are extracted from multispectral images, including contrast, entropy, and correlation. (3) Polarization characteristics: The backscattering coefficient of SAR data and the surface scattering power obtained by Yamaguchi decomposition were used. Secondary scattering power Volume scattering power Extract polarization features; Step S32: Use the improved wavelet packet transform-principal component analysis fusion algorithm to construct a multi-source data fusion model.

[0010] Preferably, a multi-source data fusion model is constructed, and the specific process is as follows: Step S321: Perform wavelet packet decomposition on the preprocessed multispectral image MS, hyperspectral image HS, and SAR polarization feature image SAR. Set the number of decomposition layers to 3 to obtain the low-frequency and high-frequency components of each data. Step S322: Apply PCA fusion to the low-frequency components and calculate the covariance matrix, eigenvalues, and eigenvectors, as shown below: ; in, It is the covariance matrix; For the first A multi-source data low-frequency component vector of 1 pixel; It is the mean vector. Total number of pixels; Indicates the transpose operation; Then, solve The eigenvector matrix is ​​obtained. and eigenvalues The first three principal components with a cumulative contribution rate of ≥95% were selected to construct a fused low-frequency component. Step S323: Adaptive weighted fusion is applied to the high-frequency components. The weights of the high-frequency components of the multispectral image MS, hyperspectral image HS, and SAR polarization feature image SAR are calculated based on their respective variances, as shown below: ; ; ; in, , and These represent the variances of the high-frequency components of multispectral, hyperspectral, and SAR data, respectively. , and The weights for the MS, HS, and SAR high-frequency components are respectively, satisfying the following conditions: ; Then, the high-frequency components are fused, as shown below: ; in, , and These are the high-frequency components of the three types of data; Step S324: Perform inverse wavelet packet transform on the fused low-frequency and high-frequency components to obtain the fused image data. .

[0011] Preferably, in step S4, a machine learning inversion model is constructed based on the fused data to invert key soil quality indicators. The specific process is as follows: Step S41: Select the core indicators of the soil quality evaluation index system; The core indicators of the soil quality evaluation index system specifically include: soil organic matter (SOM), total nitrogen (TN), available phosphorus (AP), available potassium (AK), pH value, and heavy metal cadmium (Cd) content. Step S42: Sample collection and model training; Within the target area, 50-100 sampling points were evenly distributed according to soil type and land use type. GPS positioning was used to collect topsoil samples from the 0-20cm layer. The actual values ​​of each indicator were measured in the laboratory, as shown below: ; Simultaneously, feature vectors corresponding to the fused data at the sampling points are extracted to construct a sample set (X, Y); where the feature vectors of the fused data are combinations of spectral, texture, and polarization features, denoted as... ; The number of features; Step S43: Use the Random Forest (RF) algorithm to construct a machine learning metric inversion model; Step S431, Sample set partitioning: Divide the sample set into a training set and a validation set in a 7:3 ratio; Step S432, Model parameter optimization: Optimize RF parameters, number of decision trees, maximum tree depth, and minimum number of samples for node splitting using the grid search method; Step S433, Inversion Model Construction: Using the fused data feature vector X as input and the actual soil quality index value Y as output, train the RF model as follows: ; in, For the first The prediction function of a decision tree; For the first Parameters of each decision tree; The number of decision trees; Output values ​​for the model; Step S434, Model accuracy verification: using the coefficient of determination R. 2 Root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate model accuracy; when When the value is ≥0.7, the model meets the inversion requirements; Step S44: Regional index inversion. The optimized RF model is applied to the fused data of the entire target area to obtain the spatial distribution map of each soil quality index.

[0012] Preferably, in step S5, the analytic hierarchy process (AHP)-matter-extension model is used for comprehensive evaluation of soil quality to achieve soil quality grade classification. The specific process is as follows: Step S51: Determine the weights of the evaluation indicators based on the AHP method; First, construct the hierarchical structure; The hierarchical structure includes: target layer, criteria layer, and indicator layer; Then, construct the judgment matrix; Several experts in soil environment and remote sensing were invited to conduct pairwise comparisons of the importance of indicators at the same level using the 1-9 scaling method, and a judgment matrix was constructed. As shown below: ; in, This indicates the importance of the p-th indicator relative to the q-th indicator; Indicates the matrix dimension; Finally, the weights are calculated and a consistency check is performed. Calculate the largest eigenvalue of the judgment matrix and corresponding feature vectors As shown below: ; The normalized feature vectors are the index weights. ; The consistency test formula is as follows: Consistency Indicators: ; Consistency ratio: ; in, The average random consistency index is used; when CR < 0.1, the judgment matrix meets the consistency requirement and the weights are effective. Step S52: Divide the soil quality into 5 grades: Excellent: Grade I, Good: Grade II, Average: Grade III, Poor: Grade IV, and Inferior: Grade V; Step S53, Comprehensive evaluation of matter-element extension; Step S531: Determine the matter-element matrix; Taking each pixel as the evaluation unit, let the object element of the evaluation unit be... ;in As an evaluation unit, As evaluation indicators, The value is the inversion value of the indicator; Step S532: Calculate the correlation degree of the evaluation unit to each quality level using the correlation function of extension sets. Where j = levels I-V, i = indicators 1-6; For benefit-oriented indicators SOM, TN, AP, and AK: ; For the cost-related indicator Cd: ; For interval-type index pH: ; in, , These are the lower and upper threshold values ​​for the i-th indicator at level j, respectively. Step S533: Calculate the overall correlation degree, as shown below: ; in, Let be the weight of the i-th indicator; Step S534, Grade Determination: Select the grade with the highest comprehensive correlation as the soil quality grade of the evaluation unit, i.e. The corresponding level is the evaluation result.

[0013] This invention also proposes a multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation system to realize the above-mentioned multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation method. The system includes: a data acquisition module, a data preprocessing module, a data fusion module, an index inversion module, a comprehensive evaluation module, and a result visualization module. The data acquisition module is used to acquire multi-source remote sensing data and auxiliary data, and supports automatic batch download and format conversion. The data preprocessing module integrates radiometric calibration, atmospheric correction, geometric correction, and SAR denoising functions to achieve standardized processing of multi-source data; The data fusion module has a built-in WPT-PCA fusion algorithm, which supports feature extraction and collaborative fusion of multi-source data; The indicator inversion module builds an inversion model based on the RF algorithm and supports sample training, model optimization, and regional indicator inversion. The comprehensive evaluation module integrates AHP weight calculation and matter-element extension evaluation functions, and outputs a soil quality grade distribution map. The results visualization module supports the generation, export, and interactive querying of indicator spatial distribution maps and quality grade maps.

[0014] Therefore, the present invention employs the above-mentioned multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation system and method, and the beneficial effects are as follows: (1) This invention significantly improves the evaluation accuracy by synergistic fusion of multi-source remote sensing data, complementary spectral, polarization and texture multi-dimensional information, combined with machine learning inversion model and AHP-matter extension evaluation model.

[0015] (2) This invention does not require large-scale manual sampling and laboratory analysis. Only a small number of verification samples are needed to achieve large-area soil quality evaluation. The evaluation cycle is shortened from several weeks in traditional methods to 3-5 days, which greatly improves the evaluation efficiency. It has a wide coverage and low cost, and can realize soil quality monitoring at different regional scales.

[0016] (3) Based on multi-source remote sensing data of different time phases, this invention can quickly update soil quality assessment results, capture the spatiotemporal change characteristics of soil quality, realize dynamic monitoring of soil quality, and provide technical support for early warning of soil quality degradation and pollution source tracing. (4) The evaluation system proposed in this invention integrates the functions of data processing, fusion, inversion, evaluation and visualization. It is easy to operate, supports batch processing and automated operation, and can be widely used in scenarios such as agricultural farmland protection, mining area ecological restoration and land use planning.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] Figure 1 This is a flowchart of the multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation method of the present invention; Figure 2 This is a block diagram of the multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation system of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0020] Example 1 like Figure 1 As shown, the multi-source remote sensing collaborative method for soil quality index inversion and comprehensive evaluation of this invention includes five core steps: multi-source remote sensing data acquisition, data preprocessing, data fusion, soil quality index inversion, and comprehensive soil quality evaluation. The specific implementation process is as follows: Step S1: Acquire different types of remote sensing data within the target area and construct a multi-source dataset.

[0021] (1) Multispectral remote sensing data: acquire Sentinel-2 (spatial resolution 10-60m, spectral range 490-2330nm, 13 bands) and Landsat-9 (spatial resolution 30m, spectral range 430-12500nm, 11 bands) data.

[0022] (2) Hyperspectral remote sensing data: Acquire Hyperion (spatial resolution 30m, spectral range 400-2500nm, 242 bands) or HJ-1A / 1B hyperspectral data and download them through the National Satellite Meteorological Center data platform.

[0023] (3) SAR remote sensing data: acquire Sentinel-1 C-band SAR data (spatial resolution 5-20m, dual polarization VV / VH) and download it through the ESA data sharing platform.

[0024] (4) Auxiliary data: Obtain the digital elevation model (DEM), land use type map and meteorological data (temperature and precipitation) of the target area for subsequent preprocessing and evaluation model correction.

[0025] Step S2: Preprocess the acquired multi-source remote sensing data to eliminate data noise and system errors, and ensure data consistency.

[0026] Step S21: Radiometric calibration converts the grayscale values ​​of the remote sensing image into apparent radiance or reflectance, eliminating differences in the sensor's own response.

[0027] (1) For multispectral / hyperspectral data, the apparent radiance conversion formula is as follows: ; in, Apparent radiance ( ); This is the band gain factor (provided by sensor parameters); The original grayscale value of the image; This is the band offset coefficient (provided by sensor parameters).

[0028] (2) The reflectance conversion formula is as follows: ; in, Reflectance at the top of the atmosphere; The Earth-Sun distance; The average solar irradiance at the top of the atmosphere ( ); The solar zenith angle is (°).

[0029] Step S22: Atmospheric correction is performed using the MODTRAN atmospheric radiative transfer model to eliminate the influence of atmospheric scattering and absorption on remote sensing images and obtain the true surface reflectance, as shown below: ; in, This represents the true reflectance of the Earth's surface. Atmospheric path reflectivity; The atmospheric transmittance of solar radiation to the Earth's surface; The atmospheric transmittance of radiation reflected from the Earth's surface to the sensor; The reflectivity of the atmospheric scattering layer; This refers to the hemispherical reflectance coefficient of the Earth's surface; specifically, the above parameters are calculated by inputting atmospheric profile parameters into the MODTRAN model; atmospheric profile parameters include air pressure, temperature, and humidity.

[0030] Step S23: Use the polynomial correction method to perform geometric correction to eliminate image distortion caused by terrain undulation and sensor attitude error.

[0031] First, establish the image pixel coordinates. with geographic coordinates The polynomial mapping relationship is as follows: ; ; in, , ,…and , , ... represent polynomial coefficients. These coefficients are obtained by selecting uniformly distributed ground control points within the target area, using the least squares method, and then resampling the corrected image using bilinear interpolation. At least 15 ground control points are selected using GPS measured coordinates.

[0032] Step S24: Through data resampling and registration, the multi-source remote sensing data are unified to the same spatial resolution and geographic coordinate system; the preferred spatial resolution of this invention is 10m.

[0033] Specifically, the nearest neighbor method is used to upsample low-resolution data (such as MODIS), and the bilinear interpolation method is used to downsample high-resolution data (such as Hyperion). Based on the mutual information registration algorithm, SAR data and hyperspectral data are registered with Sentinel-2 imagery as the reference, and the registration error is controlled within 0.5 pixels.

[0034] Step S25: The Lee filtering algorithm is used to denoise and polarize the SAR data to remove speckle noise, as shown below: ; in, The backscattering coefficient after denoising; The original backscattering coefficient; The noise variance (calculated from areas of the image without data); These are the filter coefficients (values ​​ranging from 1.5 to 2.0).

[0035] The Yamaguchi polarization decomposition method was used to extract SAR polarization characteristic parameters, specifically including volume scattering power. Surface scattering power and secondary scattering power As shown below: ; in, This represents the total scattered power. The noise power is obtained by solving the eigenvalues ​​of the polarization scattering matrix.

[0036] Step S3: Adopt the spectral-polarization-texture collaborative fusion strategy, combined with multi-scale decomposition and weighted fusion model, to fuse multi-source remote sensing data and achieve complementary advantages of multi-source remote sensing data.

[0037] Step S31: Feature extraction.

[0038] (1) Spectral features: Spectral features are extracted from multispectral / hyperspectral data, including vegetation index, soil index and hyperspectral absorption depth.

[0039] The vegetation indices include the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI), as shown below: ; ; in, Reflectivity in the near-infrared band; Reflectivity in the red light band; Reflectivity in the blue light band; , , and All are empirical coefficients.

[0040] Soil Index As shown below: ; in, This refers to the reflectivity in the shortwave infrared band.

[0041] High spectral absorption depth As shown below: ; in, The minimum reflectivity of the absorption valley; The continuous reflectance on both sides of the absorption valley.

[0042] (2) Texture features: Based on the gray-level co-occurrence matrix (GLCM), texture features are extracted from multispectral images, including contrast, entropy and correlation.

[0043] (3) Polarization characteristics: The backscattering coefficient of SAR data and the surface scattering power obtained by Yamaguchi decomposition were used. Secondary scattering power Volume scattering power Polarization features are extracted.

[0044] Step S32: Using the improved wavelet packet transform-principal component analysis (WPT-PCA) fusion algorithm, a multi-source data fusion model is constructed. The specific steps are as follows: Step S321: Perform wavelet packet decomposition on the preprocessed multispectral image (MS), hyperspectral image (HS), and SAR polarimetric feature image (SAR). Set the number of decomposition layers to 3 to obtain the low-frequency components (approximation coefficients) and high-frequency components (detail coefficients) of each data.

[0045] Step S322: Apply PCA fusion to the low-frequency components and calculate the covariance matrix, eigenvalues, and eigenvectors, as shown below: ; in, It is the covariance matrix; For the first A multi-source data low-frequency component vector of 1 pixel; It is the mean vector. Total number of pixels; This indicates the transpose operation.

[0046] Then, solve The eigenvector matrix is ​​obtained. and eigenvalues The first three principal components (cumulative contribution rate ≥ 95%) were selected to construct the fused low-frequency component.

[0047] Step S323: Adaptive weighted fusion is applied to the high-frequency components. The weights of the high-frequency components are calculated based on the variances of the multispectral image (MS), hyperspectral image (HS), and SAR polarimetric feature image (SAR), as shown below: ; ; ; in, , and These represent the variances of the high-frequency components of multispectral, hyperspectral, and SAR data, respectively. , and The weights for the MS, HS, and SAR high-frequency components are respectively, satisfying the following conditions: .

[0048] Then, the high-frequency components are fused, as shown below: ; in, , and These are the high-frequency components of the three types of data.

[0049] Step S324: Perform inverse wavelet packet transform on the fused low-frequency and high-frequency components to obtain the fused image data. .

[0050] Step S4: Based on fused data We constructed a machine learning inversion model to invert key indicators of soil quality.

[0051] Step S41: Select the core indicators of the soil quality evaluation index system.

[0052] The core indicators of the soil quality evaluation index system include: soil organic matter (SOM), total nitrogen (TN), available phosphorus (AP), available potassium (AK), pH value, and cadmium (Cd) content. These core indicators cover three dimensions: nutrient status, acid-base balance, and pollution level.

[0053] Step S42: Sample collection and model training.

[0054] Within the target area, 50-100 sampling points were evenly distributed according to soil type and land use type. GPS positioning was used to collect topsoil samples from the 0-20cm layer. The actual values ​​of each indicator were measured in the laboratory, as shown below: ; Simultaneously, feature vectors corresponding to the fused data at the sampling points are extracted to construct a sample set (X, Y); where the feature vectors of the fused data are combinations of spectral, texture, and polarization features, denoted as... ; The number of features.

[0055] Step S43: Use the Random Forest (RF) algorithm to construct a machine learning metric inversion model.

[0056] Step S431, Sample set division: Divide the sample set into a training set (for model training) and a validation set (for model accuracy evaluation) in a 7:3 ratio.

[0057] Step S432, Model Parameter Optimization: Optimize RF parameters using grid search method, number of decision trees ( =200), maximum tree depth ( =15), minimum number of samples for node splitting ( =5).

[0058] Step S433, Inversion Model Construction: Using the fused data feature vector X as input and the actual soil quality index value Y as output, train the RF model as follows: ; in, For the first The prediction function of a decision tree; For the first The parameters of each decision tree.

[0059] Step S434, Model accuracy verification: Using the coefficient of determination (R²) 2 The root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate model accuracy, as shown below: ; ; ; in, The actual value of the indicator; These are the model's predicted values; This is the average of the actual values; The number of samples in the validation set; when When the value is ≥0.7, the model meets the inversion requirements.

[0060] Step S44: Regional index inversion applies the optimized RF model to the fused data of the entire target region. The spatial distribution maps of various soil quality indicators were obtained.

[0061] Step S5: Use the Hierarchical Analysis-Matter-Element Extension Model (AHP) to conduct a comprehensive evaluation of soil quality and classify soil quality grades.

[0062] Step S51: Determine the weights of the evaluation indicators based on the AHP method.

[0063] First, construct a hierarchical structure.

[0064] The hierarchical structure includes: target layer (comprehensive evaluation of soil quality), criterion layer (nutrient status, acid-base balance, pollution level) and indicator layer (SOM, TN, AP, AK, pH, Cd).

[0065] Then, construct the judgment matrix.

[0066] Invite 5-8 experts in soil environment and remote sensing to conduct pairwise comparisons of the importance of indicators at the same level using the 1-9 scaling method, and construct a judgment matrix. As shown below: ; in, This indicates the importance of the p-th indicator relative to the q-th indicator; Indicates the matrix dimension.

[0067] Finally, the weights are calculated and a consistency check is performed.

[0068] Calculate the largest eigenvalue of the judgment matrix and corresponding feature vectors As shown below: ; The normalized feature vectors are the index weights. .

[0069] The consistency test formula is as follows: Consistency Indicators: ; Consistency ratio: ; in, The average random consistency index is RI = 1.24 when n = 6; when CR < 0.1, the judgment matrix meets the consistency requirement and the weights are effective.

[0070] Step S52: Divide the soil quality into 5 grades: excellent (Grade I), good (Grade II), medium (Grade III), poor (Grade IV), and inferior (Grade V). The grade threshold ranges for each indicator are shown in Table 1.

[0071] Table 1. Threshold ranges for each indicator

[0072] Step S53: Comprehensive evaluation of matter-element extension.

[0073] Step S531: Determine the matter-element matrix.

[0074] Taking each pixel as the evaluation unit, let the object element of the evaluation unit be... ;in As an evaluation unit, As evaluation indicators, This is the inversion value of the indicator.

[0075] Step S532: Calculate the correlation degree of the evaluation unit to each quality level using the correlation function of extension sets. (j = Levels I-V, i = Indicators 1-6).

[0076] For benefit-oriented indicators (SOM, TN, AP, AK): ; For cost-related indicators (Cd): ; For interval-type indicators (pH): ; in, , These are the lower and upper limits of the i-th indicator at level j, respectively.

[0077] Step S533: Calculate the overall correlation degree, as shown below: ; in, Let be the weight of the i-th indicator.

[0078] Step S534, Grade Determination: Select the grade with the highest comprehensive correlation as the soil quality grade of the evaluation unit, i.e. The corresponding level is the evaluation result.

[0079] Based on this, the present invention also proposes a multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation system, such as... Figure 2 The system shown is used to implement the above method and includes: a data acquisition module, a data preprocessing module, a data fusion module, an indicator inversion module, a comprehensive evaluation module, and a result visualization module.

[0080] The data acquisition module is used to acquire multi-source remote sensing data and auxiliary data, and supports automatic batch download and format conversion.

[0081] The data preprocessing module integrates functions such as radiometric calibration, atmospheric correction, geometric correction, and SAR denoising to achieve standardized processing of multi-source data.

[0082] The data fusion module has a built-in WPT-PCA fusion algorithm, which supports feature extraction and collaborative fusion of multi-source data.

[0083] The index inversion module builds an inversion model based on the RF algorithm, supporting sample training, model optimization, and regional index inversion.

[0084] The comprehensive evaluation module integrates AHP weight calculation and matter-element extension evaluation functions, and outputs a soil quality grade distribution map.

[0085] The results visualization module supports the generation, export, and interactive querying of indicator spatial distribution maps and quality grade maps.

[0086] Example 2 This embodiment uses a farmland area in the Northeast Plain as the research object to evaluate soil quality.

[0087] 1. Overview of the implementation area.

[0088] The target area is located in a farmland area of ​​the Songnen Plain in Northeast China, covering an area of ​​approximately 600 km². 2 The land use type is mainly corn-soybean rotation farmland, the soil type is black soil (with high background organic matter content), and the terrain is flat (elevation 120-180m). It is necessary to evaluate the current status of black soil quality in this area to provide a basis for black soil protection and farmland fertility improvement.

[0089] 2. Data acquisition.

[0090] Multispectral data: Sentinel-2 image (August 2024, corn grain-filling stage, cloud cover <3%), 10m spatial resolution, with 7 bands selected: B2 (490nm), B3 (560nm), B4 (665nm), B5 (705nm), B8 (842nm), B11 (1610nm), and B12 (2190nm).

[0091] Hyperspectral data: HJ-1A hyperspectral image (August 2024, time difference with Sentinel-2 ≤ 5 days), 30m spatial resolution, 120 noise-free bands in the range of 400-2500nm were selected (with a focus on retaining the sensitive bands of black soil organic matter).

[0092] SAR data: Sentinel-1 C-band imagery (August 2024, VV / VH dual polarization, 10m spatial resolution).

[0093] Supporting data: SRTM DEM (30m), regional land use map (1:50,000), meteorological data for August 2024 (average temperature 24℃, precipitation 120mm).

[0094] Measured data: In August 2024, 70 sampling points were simultaneously set up (distributed according to the black soil thickness gradient), and topsoil samples from 0-20cm were collected. Laboratory-measured indicators: SOM (15.2-42.8g / kg), TN (1.05-2.86g / kg), AP (8.2-50.3mg / kg), AK (85-260mg / kg), pH (5.6-7.8), Cd (0.03-0.65mg / kg).

[0095] 3. Data preprocessing.

[0096] In response to the high humidity characteristics of the Northeast Plain in summer, atmospheric correction has added a water vapor correction module: Input the regional water vapor content (2.5 g / cm³) into the MODTRAN model. 2 This improves the accuracy of surface reflectivity.

[0097] Geometric correction was performed using 25 ground control points (GCPs) and a quadratic polynomial correction method, resulting in an RMSE of 0.25 pixels after correction.

[0098] SAR data processing: Lee filtering is used ( =1.6) Remove speckle noise, Yamaguchi polarization decomposition extraction , , feature.

[0099] Data registration: The hyperspectral data was downsampled to 10m and registered with the Sentinel-2 and SAR data to the WGS84 coordinate system.

[0100] 4. Data fusion.

[0101] Feature extraction: Supplementing black soil-specific spectral features (absorption area of ​​black soil organic matter sensitive bands), and combining multispectral NDVI, SI, hyperspectral absorption depth, and SAR polarization features, a total of 22 feature variables were extracted.

[0102] WPT-PCA fusion: After 3-level wavelet packet decomposition, the low-frequency components are fused using PCA (cumulative contribution rate of 97.2%), and the high-frequency components are fused based on variance weighting to obtain fused data.

[0103] 5. Indicator inversion.

[0104] Sample splitting: The 70 sampling points were divided into a training set (49 points) and a validation set (21 points) in a 7:3 ratio.

[0105] RF model optimization: Considering the high organic matter content of black soil, the weights of SOM-sensitive features are increased, and the model parameters are adjusted. =220, =16; Accuracy verification: Inversion accuracy of each index: SOM(R) 2 =0.85, RMSE=2.1g / kg), TN (R 2 =0.81, RMSE=0.11g / kg), AP (R 2 =0.79, RMSE=2.2mg / kg), AK (R 2 =0.82, RMSE=14.8mg / kg), pH (R 2 =0.77, RMSE=0.28), Cd(R 2 =0.83, RMSE=0.04mg / kg), which meets the accuracy requirements.

[0106] 6. Overall evaluation.

[0107] AHP weights: Based on the key points of black soil protection, the weights were adjusted to SOM (0.30), TN (0.20), AP (0.15), AK (0.15), pH (0.10), and Cd (0.10). The consistency test CR = 0.07 < 0.1.

[0108] Quality grade distribution: Excellent (Grade I) accounted for 28.5% (SOM>30g / kg), Good (Grade II) accounted for 42.3%, Medium (Grade III) accounted for 24.1%, Poor (Grade IV) accounted for 5.1%, and Inferior (Grade V) accounted for 0%.

[0109] Results analysis: The poor-quality areas were concentrated in plots that had been continuously cropped with maize for a long time (SOM<15g / kg), which verified the pattern that "continuous cropping leads to a decrease in organic matter in black soil" and provided a basis for subsequent crop rotation adjustments.

[0110] Therefore, this invention employs the aforementioned multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation system and method. By collaboratively fusing multi-source remote sensing data and complementing multi-dimensional information such as spectral, polarization, and texture, and combining machine learning inversion models with AHP-matter-extension evaluation models, the evaluation accuracy is significantly improved. Large-scale manual sampling and laboratory analysis are unnecessary; only a small number of validation samples are required to achieve large-area soil quality evaluation. The evaluation cycle is shortened from several weeks using traditional methods to 3-5 days, resulting in a significant increase in evaluation efficiency. It also offers wide coverage and low cost, enabling soil quality monitoring at different regional scales.

[0111] This invention rapidly updates soil quality assessment results based on multi-source remote sensing data from different time phases, captures the spatiotemporal variation characteristics of soil quality, and realizes dynamic monitoring of soil quality, providing technical support for early warning of soil quality degradation and source tracing of pollution. The proposed assessment system integrates the entire process of data processing, fusion, inversion, evaluation, and visualization. It is easy to operate, supports batch processing and automated operation, and can be widely used in scenarios such as agricultural farmland protection, mining area ecological restoration, and land use planning.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for soil quality index inversion and comprehensive evaluation based on multi-source remote sensing, characterized in that, Includes the following steps: Step S1: Acquire different types of remote sensing data within the target area and construct a multi-source dataset; Step S2: Preprocess the acquired multi-source remote sensing data to eliminate data noise and system errors, and ensure data consistency; Step S3: Employ a spectral-polarization-texture collaborative fusion strategy, combined with a multi-scale decomposition and weighted fusion model, to fuse multi-source remote sensing data; Step S4: Based on the fused data, construct a machine learning inversion model to invert key soil quality indicators; Step S5: Use the analytic hierarchy process (AHP)-matter extension model to conduct a comprehensive evaluation of soil quality and classify soil quality grades.

2. The method for inverting and comprehensively evaluating soil quality indicators using multi-source remote sensing synergy as described in claim 1, characterized in that, In step S1, different types of remote sensing data within the target area are acquired to construct a multi-source data set, specifically including: multispectral remote sensing data, hyperspectral remote sensing data, SAR remote sensing data, and auxiliary data; The auxiliary data includes: digital elevation models of the target area, land use type maps, and meteorological data.

3. The method for inverting and comprehensively evaluating soil quality indicators using multi-source remote sensing synergy according to claim 1, characterized in that, In step S2, the acquired multi-source remote sensing data is preprocessed to eliminate data noise and systematic errors, ensuring data consistency. The specific process is as follows: Step S21: Radiometric calibration converts the grayscale values ​​of the remote sensing image into apparent radiance or reflectance, eliminating differences in the sensor's own response. Step S22: Use the MODTRAN atmospheric radiative transfer model to perform atmospheric correction, eliminate the influence of atmospheric scattering and absorption on remote sensing images, and obtain the true surface reflectance. Step S23: Use the polynomial correction method to perform geometric correction to eliminate image distortion caused by terrain undulation and sensor attitude error; Establish image pixel coordinates with geographic coordinates The polynomial mapping relationship is used to solve the coefficients using the least squares method, and the corrected image is resampled using bilinear interpolation. Step S24: Through data resampling and registration, unify the multi-source remote sensing data to the same spatial resolution and geographic coordinate system; The nearest neighbor method is used to upsample low-resolution data, and the bilinear interpolation method is used to downsample high-resolution data. Based on the mutual information registration algorithm, SAR data and hyperspectral data are registered with Sentinel-2 imagery as the reference, and the registration error is controlled within 0.5 pixels. Step S25: Use the Lee filtering algorithm to denoise and polarize the SAR data to remove speckle noise from the SAR data; The Yamaguchi polarization decomposition method was used to extract SAR polarization characteristic parameters, specifically including volume scattering power. Surface scattering power and secondary scattering power .

4. The method for inverting and comprehensively evaluating soil quality indicators using multi-source remote sensing synergy as described in claim 1, characterized in that, In step S3, a spectral-polarization-texture collaborative fusion strategy is adopted, combined with a multi-scale decomposition and weighted fusion model, to fuse multi-source remote sensing data and achieve complementary advantages of multi-source remote sensing data. The specific process is as follows: Step S31: Feature extraction; (1) Spectral features: Spectral features are extracted from multispectral / hyperspectral data, including vegetation index, soil index and hyperspectral absorption depth; (2) Texture features: based on Gray-level co-occurrence matrix (GLCM) extracts texture features from multispectral images, including contrast, entropy, and correlation. (3) Polarization characteristics: The backscattering coefficient of SAR data and the surface scattering power obtained by Yamaguchi decomposition were used. Secondary scattering power Volume scattering power Extract polarization features; Step S32: Use the improved wavelet packet transform-principal component analysis fusion algorithm to construct a multi-source data fusion model.

5. The method for inverting and comprehensively evaluating soil quality indicators using multi-source remote sensing synergy according to claim 4, characterized in that, The specific process for constructing a multi-source data fusion model is as follows: Step S321: Perform wavelet packet decomposition on the preprocessed multispectral image MS, hyperspectral image HS, and SAR polarization feature image SAR. Set the number of decomposition layers to 3 to obtain the low-frequency and high-frequency components of each data. Step S322: Apply PCA fusion to the low-frequency components and calculate the covariance matrix, eigenvalues, and eigenvectors, as shown below: ； in, It is the covariance matrix; For the first A multi-source data low-frequency component vector of 1 pixel; It is the mean vector. Total number of pixels; Indicates the transpose operation; Then, solve The eigenvector matrix is ​​obtained. and eigenvalues The first three principal components with a cumulative contribution rate of ≥95% were selected to construct a fused low-frequency component. Step S323: Adaptive weighted fusion is applied to the high-frequency components. The weights of the high-frequency components of the multispectral image MS, hyperspectral image HS, and SAR polarization feature image SAR are calculated based on their respective variances, as shown below: ； ； ； in, , and These represent the variances of the high-frequency components of multispectral, hyperspectral, and SAR data, respectively. , and The weights for the MS, HS, and SAR high-frequency components are respectively, satisfying the following conditions: ; Then, the high-frequency components are fused, as shown below: ； in, , and These are the high-frequency components of the three types of data; Step S324: Perform inverse wavelet packet transform on the fused low-frequency and high-frequency components to obtain the fused image data. .

6. The method for inverting and comprehensively evaluating soil quality indicators using multi-source remote sensing synergy according to claim 1, characterized in that, In step S4, a machine learning inversion model is constructed based on the fused data to invert key soil quality indicators. The specific process is as follows: Step S41: Select the core indicators of the soil quality evaluation index system; The core indicators of the soil quality evaluation index system specifically include: soil organic matter (SOM), total nitrogen (TN), available phosphorus (AP), available potassium (AK), pH value, and heavy metal cadmium (Cd) content. Step S42: Sample collection and model training; Within the target area, 50-100 sampling points were evenly distributed according to soil type and land use type. GPS positioning was used to collect topsoil samples from the 0-20cm layer. The actual values ​​of each indicator were measured in the laboratory, as shown below: ； Simultaneously, feature vectors corresponding to the fused data at the sampling points are extracted to construct a sample set (X, Y); where the feature vectors of the fused data are combinations of spectral, texture, and polarization features, denoted as... ; The number of features; Step S43: Use the Random Forest (RF) algorithm to construct a machine learning metric inversion model; Step S431, Sample set partitioning: Divide the sample set into a training set and a validation set in a 7:3 ratio; Step S432, Model parameter optimization: Optimize RF parameters, number of decision trees, maximum tree depth, and minimum number of samples for node splitting using the grid search method; Step S433, Inversion Model Construction: Using the fused data feature vector X as input and the actual soil quality index value Y as output, train the RF model as follows: ； in, For the first The prediction function of each decision tree; For the first Parameters of each decision tree; The number of decision trees; Output values ​​for the model; Step S434, Model accuracy verification: using the coefficient of determination R. 2 Root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate model accuracy; when When the value is ≥0.7, the model meets the inversion requirements; Step S44: Regional index inversion. The optimized RF model is applied to the fused data of the entire target area to obtain the spatial distribution map of each soil quality index.

7. The method for inverting and comprehensively evaluating soil quality indicators using multi-source remote sensing synergy according to claim 1, characterized in that, In step S5, the analytic hierarchy process (AHP)-matter-extension model is used to comprehensively evaluate soil quality and classify soil quality grades. The specific process is as follows: Step S51: Determine the weights of the evaluation indicators based on the AHP method; First, construct the hierarchical structure; The hierarchical structure includes: target layer, criteria layer, and indicator layer; Then, construct the judgment matrix; Several experts in soil environment and remote sensing were invited to conduct pairwise comparisons of the importance of indicators at the same level using the 1-9 scaling method, and a judgment matrix was constructed. As shown below: ； in, This indicates the importance of the p-th indicator relative to the q-th indicator; Indicates the matrix dimension; Finally, the weights are calculated and a consistency check is performed. Calculate the largest eigenvalue of the judgment matrix and corresponding feature vectors As shown below: ； The normalized feature vectors are the index weights. ; The consistency test formula is as follows: Consistency Indicators: ; Consistency ratio: ; in, The average random consistency index is used; when CR < 0.1, the judgment matrix meets the consistency requirement and the weights are effective. Step S52: Divide the soil quality into 5 grades: Excellent: Grade I, Good: Grade II, Average: Grade III, Poor: Grade IV, and Inferior: Grade V; Step S53, Comprehensive evaluation of matter-element extension; Step S531: Determine the matter-element matrix; Taking each pixel as the evaluation unit, let the object element of the evaluation unit be... ;in As an evaluation unit, As evaluation indicators, The value is the inversion value of the indicator; Step S532: Calculate the correlation degree of the evaluation unit to each quality level using the correlation function of extension sets. Where j = levels I-V, i = indicators 1-6; For benefit-oriented indicators SOM, TN, AP, and AK: ； For the cost-related indicator Cd: ； For interval-type index pH: ； in, , These are the lower and upper threshold values ​​for the i-th indicator at level j, respectively. Step S533: Calculate the overall correlation degree, as shown below: ； in, Let be the weight of the i-th indicator; Step S534, Grade Determination: Select the grade with the highest comprehensive correlation as the soil quality grade of the evaluation unit, i.e. The corresponding level is the evaluation result.

8. A multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation system, used to implement the multi-source remote sensing collaborative soil quality index inversion and comprehensive evaluation method described in any one of claims 1-7, characterized in that, The system includes: a data acquisition module, a data preprocessing module, a data fusion module, an indicator inversion module, a comprehensive evaluation module, and a result visualization module; The data acquisition module is used to acquire multi-source remote sensing data and auxiliary data, and supports automatic batch download and format conversion. The data preprocessing module integrates radiometric calibration, atmospheric correction, geometric correction, and SAR denoising functions to achieve standardized processing of multi-source data; The data fusion module has a built-in WPT-PCA fusion algorithm, which supports feature extraction and collaborative fusion of multi-source data; The indicator inversion module builds an inversion model based on the RF algorithm and supports sample training, model optimization, and regional indicator inversion. The comprehensive evaluation module integrates AHP weight calculation and matter-element extension evaluation functions, and outputs a soil quality grade distribution map. The results visualization module supports the generation, export, and interactive querying of indicator spatial distribution maps and quality grade maps.

Images