Soil salinity inversion method, system and equipment based on multi-modal remote sensing data fusion and storage medium

Abstract

The invention discloses a soil salinity inversion method, system and device based on multi-mode remote sensing data fusion and a storage medium, and is applied to the technical field of soil quality monitoring, and the method comprises the steps: obtaining radar and optical remote sensing data of a to-be-inverted region of soil salinity, and carrying out the conductivity measurement, and taking the obtained data as an inversion index of the soil salinity; obtaining multi-modal remote sensing features based on radar and optical remote sensing data fusion, optimizing a remote sensing feature combination by combining the radar and optical remote sensing data and based on a random forest algorithm and utilizing a recursive feature elimination method, and performing model performance evaluation on the remote sensing feature combination by adopting cross validation to obtain an optimal remote sensing feature combination; and based on the optimal remote sensing feature combination, constructing and training a soil conductivity prediction model based on a random forest algorithm, and inputting to-be-measured data to the conductivity prediction model after hyper-parameter adjustment to complete soil salinity inversion. According to the method, more accurate inversion of the soil salinity under the interaction influence of a complex environment and human factors is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of soil quality monitoring, and in particular to a soil salinity inversion method, system, equipment and storage medium based on multimodal remote sensing data fusion. Background Art

[0002] Soil salinization is a serious environmental challenge, particularly in arid and semi-arid regions, severely impacting agricultural productivity, ecosystem health, and water resource management. Globally, approximately 831 million hectares of land are affected by soil salinization, a large portion of which is located in irrigated agricultural areas. Salinization occurs when excessive salt accumulates in the soil, impairing plant growth and soil fertility. This process is often exacerbated by natural factors such as inadequate irrigation practices, poor drainage, and high evaporation rates. As agriculture remains the backbone of many economies, monitoring and mitigating soil salinization is crucial to ensuring long-term food security and sustainable land use.

[0003] The Hetao Irrigation District (HID), located in China's Inner Mongolia Autonomous Region, is one of the most important agricultural production areas in northern China. However, due to its flat terrain, high evaporation, and intensive irrigation, the HID has long been plagued by soil salinization. Salt accumulation in the region not only reduces crop yields but also threatens the long-term sustainability of agricultural production. Traditional soil salinization monitoring methods, such as ground sampling and laboratory analysis, while accurate, are labor-intensive, time-consuming, and space-constrained. Therefore, more efficient, scalable, and accurate methods are urgently needed to monitor soil salinization over large areas. Remote sensing technology has emerged as a promising tool for soil salinization monitoring, providing a non-invasive, cost-effective, and scalable method to assess soil conditions over large geographic areas. In particular, satellite-based remote sensing provides valuable data for detecting and monitoring salinization-induced changes in soil and vegetation cover. Various studies have utilized optical and radar satellite data to monitor soil salinity, with success depending on the type of data, spatial resolution, and analytical techniques employed. Optical data are often used to derive vegetation indices such as the Normalized Difference Vegetation Index (NDVI) and the Salinity Index (SI), which indirectly indicate soil salt through the detected vegetation stress. Radar data can provide information on soil moisture and surface roughness, which are also related to salinity.

[0004] However, despite these advances, accurately retrieving soil salinity using remote sensing data remains challenging, especially in regions such as the HID, where soil salinity is influenced by a complex interaction of natural and anthropogenic factors. Combining multiple remote sensing data sources, such as optical and radar imagery, has the potential to improve the accuracy of salinity monitoring by leveraging the strengths of each data type. For example, optical data can capture spectral reflectance that is associated with vegetation stress and soil properties, while radar data can penetrate clouds and provide consistent measurements of soil surface conditions independent of weather conditions.

[0005] Therefore, how to provide a soil salinity inversion method, system, equipment and storage medium based on multimodal remote sensing data fusion that can achieve more accurate inversion of soil salinity under the influence of complex environments and the interaction of human factors is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention proposes a soil salinity inversion method, system, device and storage medium based on multimodal remote sensing data fusion.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A soil salinity inversion method based on multimodal remote sensing data fusion, including:

[0009] Step 1: Obtain radar remote sensing data and optical remote sensing data of the area where soil salinity is to be inverted, and measure the soil electrical conductivity as an inversion indicator of soil salinity;

[0010] Step 2: Obtain multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data. Combine the radar remote sensing data with the optical remote sensing data, and optimize the remote sensing feature combination using the recursive feature elimination method based on the random forest algorithm. Use cross-validation to evaluate the model performance of the optimized remote sensing feature combination to obtain the optimal remote sensing feature combination.

[0011] Step 3: Based on the optimal remote sensing feature combination, a soil conductivity prediction model based on the random forest algorithm is constructed and trained. The test data is input into the soil conductivity prediction model after hyperparameter adjustment to obtain soil conductivity and complete soil salinity inversion.

[0012] Optionally, in step 1, the radar remote sensing data includes: VV band and VH band.

[0013] Optionally, in step 1, the optical remote sensing data includes: blue band B2, green band B3, red band B4, first red edge band B5, second red edge band B6, third red edge band B7, near infrared band B8, fourth red edge band B8A, first shortwave infrared band B11 and second shortwave infrared band B12.

[0014] Optionally, in step 2, the multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data include: normalized difference vegetation index NDVI, enhanced vegetation index EVI, soil adjusted vegetation index SAVI, modified soil adjusted vegetation index MSAVI, slope adjusted vegetation index TSAVI, adjusted normalized vegetation index OSAVI, adaptive soil adjusted vegetation index ATSAVI, vegetation index PVI, near infrared vegetation index NIRv, chlorophyll index MTCI, water sensitive vegetation index SSAI, chromaticity index Clr, vegetation health index CIg, first normalized difference index NDRE1, second normalized difference index NDRE2, first normalized difference vegetation index NDVIr1, second normalized difference vegetation index NDVIr2, third normalized difference vegetation index NDVIr3, radar vegetation index VV / VH Ratio, radar vegetation index RVI, dual-polarization SAR vegetation index DPSVI, normalized difference polarimetric index NDPI, a spectral transformation method Brightness, greenness index Greenness and wetness index Wetness.

[0015] Optionally, in step 2, the remote sensing feature combination is optimized using a recursive feature elimination method based on the random forest algorithm, specifically:

[0016] In each iteration, the importance score of the feature is calculated by using the Gini index, and the feature with the least contribution to the model is eliminated;

[0017] The Gini index is an indicator for measuring impurity. For a node t, the Gini index is as follows:

[0018]

[0019] Where C is the total number of categories; p i is the proportion of samples belonging to category i to the total samples of the node;

[0020] The smaller the Gini index, the purer the samples in the node, which means that most samples belong to the same category;

[0021] When a feature is used to split a node, the data is divided into two child nodes, left and right. The Gini gain measures the reduction in the Gini index after the split, as follows:

[0022]

[0023] Among them, Gini(t) is the Gini index of the node before splitting; Gini(t left ) and Gini(t right ) are the Gini indexes of the left and right child nodes after splitting; N left and N right are the number of samples of the left and right child nodes after splitting; N is the total number of samples of the current node;

[0024] The larger the Gini gain, the greater the contribution of the feature to the classification;

[0025] In a random forest, each split node of each tree calculates the Gini gain based on a certain feature. The importance score of the feature is the cumulative average of the Gini gains of all trees in the forest, as follows:

[0026]

[0027] Where T is the total number of trees in the random forest; is the Gini gain of feature j in one split in the k-th tree;

[0028] Random Forest evaluates the importance of each feature using the above formula and performs recursive feature elimination.

[0029] Optionally, in step 2, cross-validation is used to evaluate the model performance of the optimized remote sensing feature combination, specifically:

[0030] The 5-fold cross-validation method was used to randomly divide the sample data into 5 mutually exclusive subsets. Four of the subsets were used as training sets each time, and the remaining subset was used as the test set for a total of 5 iterations.

[0031] In each iteration, the random forest model is used to train the training set, and the trained model is used to predict the test set to obtain the predicted value and the true value. The average determination coefficient and root mean square error of 5 iterations are used to evaluate the model performance of different remote sensing feature combinations.

[0032] Optionally, in step 3, hyperparameters are adjusted, including: the number of trees in the random forest ntrees, the maximum number of nodes per tree maxnodes, the minimum number of samples of the terminal node nodesize, and the number of samples drawn from the training set for each tree sampsize.

[0033] The present invention also provides a soil salinity inversion system based on multimodal remote sensing data fusion using a soil salinity inversion method based on multimodal remote sensing data fusion, comprising:

[0034] Data acquisition module: used to obtain radar remote sensing data and optical remote sensing data of the area where soil salinity is to be inverted, and to measure the electrical conductivity of the soil as an inversion indicator of soil salinity;

[0035] Feature screening module: used to obtain multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data, and to optimize the remote sensing feature combination based on the random forest algorithm and the recursive feature elimination method based on the combination of radar remote sensing data and optical remote sensing data. The model performance of the optimized remote sensing feature combination is evaluated by cross-validation to obtain the optimal remote sensing feature combination.

[0036] Salinity inversion module: used to construct and train a soil conductivity prediction model based on the random forest algorithm based on the optimal remote sensing feature combination, input the test data into the soil conductivity prediction model after hyperparameter adjustment, obtain soil conductivity, and complete soil salinity inversion.

[0037] The present invention further provides an electronic device, comprising:

[0038] memory for storing computer programs;

[0039] The processor is configured to implement the steps of a soil salinity inversion method based on multimodal remote sensing data fusion when executing the computer program.

[0040] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements the steps of a soil salinity inversion method based on multimodal remote sensing data fusion.

[0041] It can be seen from the above technical solutions that compared with the prior art, the present invention proposes a soil salinity inversion method, system, device and storage medium based on multimodal remote sensing data fusion. The present invention integrates two complementary Sentinel-1 radar data (providing insights into soil moisture and surface characteristics) and Sentinel-2 optical data (capturing the spectral characteristics of vegetation and soil related to salinity), extracts multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data, and optimizes and evaluates the feature combination through recursive feature elimination and cross-validation methods, and obtains the optimal feature combination (Greenness, Wetness, VV, B11 and NDVI) suitable for the most typical Hetao Irrigation District. Finally, the hyperparameter-adjusted model is applied to the soil salinity inversion in the Hetao Irrigation District in 2021, proving the accuracy of the present invention. In summary, the present invention achieves a more accurate inversion of soil salinity under the influence of complex environments and the interaction of human factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0043] Figure 1 Schematic diagram of the method of the present invention.

[0044] Figure 2 This is a schematic diagram of the distribution of soil sample points in the Hetao Irrigation District in 2021 of the present invention.

[0045] Figure 3 Schematic diagram of the relationship between the number of variables and cross-validation RMSE based on the random forest regression model of the present invention.

[0046] Figure 4 This is a schematic diagram of the importance ranking of characteristic variables of the present invention.

[0047] Figure 5 This is a schematic diagram of the relationship between the actual value and predicted value of EC in 2021 of the present invention.

[0048] Figure 6 This is a schematic diagram of the salinization degree classification of cultivated land in the Hetao Irrigation District in 2021 according to the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] Example 1:

[0051] Example 1 of the present invention discloses a soil salinity inversion method based on multimodal remote sensing data fusion, such as Figure 1 Shown, including:

[0052] Step 1: Obtain radar remote sensing data and optical remote sensing data of the area where soil salinity is to be inverted, and measure the soil electrical conductivity as an inversion indicator of soil salinity.

[0053] Radar remote sensing data, including VV band and VH band.

[0054] Radar echo signals in the VV band (vertical-vertical polarization) are sensitive to soil moisture and surface roughness, and can provide soil moisture information in salinized areas. In salinized areas, the VV band can effectively reveal the wetness of saline soil, thereby reflecting the degree of salinization.

[0055] The VH band (vertical-horizontal polarization) helps identify the unique surface characteristics of salinized soils by reflecting the roughness and structure of the soil surface. In areas with severe salinization, the soil surface typically becomes hard and rough, and the VH band can capture these changes, thereby improving the accuracy of salinized soil identification.

[0056] Optical remote sensing data, including: blue band B2 (Blue), green band B3 (Green), red band B4 (Red), first red edge band B5 (Red Edge1), second red edge band B6 (Red Edge2), third red edge band B7 (RedEdge3), near-infrared band B8 (NIR), fourth red edge band B8A (Red Edge4), first shortwave infrared band B11 (SWIR1), and second shortwave infrared band B12 (SWIR2).

[0057] Step 2: Obtain multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data, and combine radar remote sensing data with optical remote sensing data. Based on the random forest algorithm, the recursive feature elimination method (Recursive Feature Elimination) is used to optimize the remote sensing feature combination. The model performance of the optimized remote sensing feature combination is evaluated using cross-validation to obtain the optimal remote sensing feature combination.

[0058] The multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data were selected based on their ability to effectively capture key information closely related to soil salinization, such as vegetation conditions, surface water distribution, brightness characteristics, and humidity. The specific selection criteria are as follows:

[0059] (1) Vegetation Index: This index indirectly reflects the impact of soil salinization on vegetation by quantifying vegetation growth, coverage, and health. Salinized soils usually inhibit vegetation growth, resulting in a significant decrease in the vegetation index, which in turn provides a reliable indirect indicator for the inversion of soil salinization.

[0060] (2) Moisture index: Soil salinization significantly affects the water retention capacity of the soil. These indices can better characterize the surface and soil moisture characteristics, providing key support for the extraction of moisture-related variables in the inversion, thereby improving the inversion accuracy.

[0061] (3) Spectral index: By selecting sensitive spectral bands, it is possible to capture the significant changes in the spectral characteristics of salinized soil and vegetation, especially the changes in spectral reflectance caused by salinization, providing rich spectral information for the quantitative analysis of the degree of soil salinization.

[0062] (4) Radar characteristics: Radar images are highly sensitive to moisture and surface structure, especially in arid and salinized areas. They can supplement the soil moisture and surface roughness information that cannot be obtained by optical images, thereby enhancing the monitoring capability of salinized soil.

[0063] (5) Pyramid transformation features: The Pyramid transformation method based on multispectral imagery extracts brightness, greenness, and humidity features, reflecting surface characteristics from a comprehensive perspective. These features effectively distinguish the spectral differences between salinized soil and other soil types, improving the reliability and applicability of the inversion.

[0064] In summary, these features cover the multi-dimensional information of optical and radar images, which helps to comprehensively assess the degree of soil salinization and provide accurate inversion results, thus providing a scientific basis for regional soil salinization management and improvement strategy formulation.

[0065] Therefore, the multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data are shown in Table 1, including: normalized difference vegetation index NDVI, enhanced vegetation index EVI, soil adjusted vegetation index SAVI, modified soil adjusted vegetation index MSAVI, slope adjusted vegetation index TSAVI, adjusted normalized vegetation index OSAVI, adaptive soil adjusted vegetation index ATSAVI, vegetation index PVI, near infrared vegetation index NIRv, chlorophyll index MTCI, water sensitive vegetation index SSAI, chromaticity index Clr, vegetation health index CIg, first normalized difference index NDRE1, second normalized difference index NDRE2, first normalized difference vegetation index NDVIr1, second normalized difference vegetation index NDVIr2, third normalized difference vegetation index NDVIr3, radar vegetation index VV / VH Ratio, radar vegetation index RVI, dual-polarization SAR vegetation index DPSVI, normalized difference polarimetric index NDPI, a spectral transformation method Brightness, greenness index Greenness and wetness index Wetness.

[0066] Table 1 Multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data

[0067]

[0068]

[0069]

[0070]

[0071] Based on the random forest algorithm, the recursive feature elimination method is used to optimize the remote sensing feature combination, specifically:

[0072] In each iteration, the Gini index was used to calculate the importance score of the features and the features with the smallest contribution to the model were eliminated. For example, the slope aspect and soil classification features had importance scores of 0.01 and 0.03, respectively, in the fifth iteration. They were considered to have limited explanatory power for the target variable and were ultimately eliminated.

[0073] The Gini index is an indicator for measuring impurity. For a node t, the Gini index is as follows:

[0074]

[0075] Where C is the total number of categories; p i is the proportion of samples belonging to category i to the total samples of the node;

[0076] The smaller the Gini index, the purer the samples in the node, which means that most samples belong to the same category;

[0077] When a feature is used to split a node, the data is divided into two child nodes, left and right. The Gini gain measures the reduction in the Gini index after the split, as follows:

[0078]

[0079] Among them, Gini(t) is the Gini index of the node before splitting; Gini(t left ) and Gini(t right ) are the Gini indexes of the left and right child nodes after splitting; N left and N right are the number of samples of the left and right child nodes after splitting; N is the total number of samples of the current node;

[0080] The larger the Gini gain, the greater the contribution of the feature to the classification;

[0081] In a random forest, each split node of each tree calculates the Gini gain based on a certain feature. The importance score of the feature is the cumulative average of the Gini gains of all trees in the forest, as follows:

[0082]

[0083] Where T is the total number of trees in the random forest; is the Gini gain of feature j in one split in the k-th tree;

[0084] Random Forest evaluates the importance of each feature using the above formula and performs recursive feature elimination.

[0085] Cross-validation is used to evaluate the model performance of the optimized remote sensing feature combination, specifically:

[0086] The 5-fold cross-validation method was used to randomly divide the sample data into 5 mutually exclusive subsets. Four of the subsets were used as training sets each time, and the remaining subset was used as the test set for a total of 5 iterations.

[0087] In each iteration, the random forest model is used to train the training set, and the trained model is used to predict the test set to obtain the predicted value and the true value. The average determination coefficient and root mean square error of 5 iterations are used to evaluate the model performance of different remote sensing feature combinations.

[0088] Step 3: Based on the optimal remote sensing feature combination, a soil conductivity prediction model based on the random forest algorithm is constructed and trained. The test data is input into the soil conductivity prediction model after hyperparameter adjustment to obtain soil conductivity and complete soil salinity inversion.

[0089] Hyperparameter adjustment includes: the number of trees in the random forest ntrees, the maximum number of nodes per tree maxnodes, the minimum number of samples of the terminal node nodesize, and the number of samples sampsize drawn from the training set for each tree.

[0090] Example 2:

[0091] Example 2 of the present invention discloses a specific application of a soil salinity inversion method based on multimodal remote sensing data fusion, as follows:

[0092] Take the Hetao Irrigation District (HID), a typical example. Located in China's Inner Mongolia Autonomous Region, HID is one of the largest irrigation areas in Asia and one of the most important agricultural production regions. Covering approximately 574,000 hectares, the region plays a vital role in China's agricultural production, particularly in the cultivation of staple crops such as wheat and corn. However, the HID is located in an arid region with very low annual precipitation (150-200 mm) and extremely high evapotranspiration (up to 2000 mm), resulting in significant water stress. The HID primarily relies on water from the Yellow River for irrigation, but inappropriate irrigation practices and an inadequate drainage system have led to widespread secondary salinization. Secondary salinization occurs when groundwater, carrying salt, rises to the surface due to overirrigation and capillary action, leaving behind salt after evaporation. The flat terrain and inefficient drainage system further exacerbate salt accumulation, deteriorating soil quality and reducing agricultural productivity. As a key agricultural production base in northern China, the sustainability of the HID is threatened by soil salinization. Therefore, developing precise monitoring and management strategies to mitigate this problem and ensure the viability of agriculture in the region is urgent.

[0093] Radar remote sensing data and optical remote sensing data are obtained from Sentinel-1 and Sentinel-2 data respectively.

[0094] Sentinel-1's Synthetic Aperture Radar (SAR) data provide crucial insights into surface dynamics in the Hetao Irrigation Area, particularly the direct impacts of soil moisture and roughness on salinization. Sentinel-1's C-band dual-polarization capability (VV and VH) offers unparalleled advantages in tracking changes in soil texture and moisture, providing data even in the absence of visible light or during cloudy weather conditions. Unlike optical data, Sentinel-1 can penetrate vegetation, capturing surface features critical for salinization assessment, making it suitable for salinization monitoring in both cultivated and non-cultivated lands. Its 12-day temporal resolution ensures frequent updates, enabling the monitoring of seasonal variations in soil moisture and salt transport on a detailed timescale. This high revisit frequency makes the radar data almost "pulse-like," providing a consistent temporal record of soil conditions. When interpreted in conjunction with environmental variables such as irrigation practices and precipitation, the Sentinel-1 data archive from 2017 to 2023 provides users with a time series perspective, helping them understand how human activities and natural factors influence salinization patterns in the region. The 10-meter spatial resolution can capture fine-scale heterogeneity, especially in transition zones where salinity increases or decreases due to changes in land management or natural drainage.

[0095] Complementing the capabilities of the Sentinel-1 radar, Sentinel-2 provides a detailed spectral analysis of the Hetao Irrigation Area's surface using its 13 multispectral bands. Sentinel-2's ability to identify subtle changes in vegetation health and soil reflectance is crucial for understanding how salinization manifests itself in the landscape. Sentinel-2 data is particularly valuable in areas where salinization is impacting crop growth, as its spectral bands, including visible, near-infrared (NIR), and shortwave infrared (SWIR), can capture vegetation stress, water changes, and the formation of salt crusts. By leveraging Sentinel-2's five-day revisit cycle, the present invention ensures that temporal trends in vegetation and bare soil can be monitored, accurately correlating remotely sensed indices such as NDVI and the Salinity Index (SI) with field-measured electrical conductivity (EC). This frequent monitoring ensures that rapid changes in crop health during critical growing seasons can be observed, identifying areas where salinization may be hindering agricultural productivity.

[0096] By combining Sentinel-1 radar data with Sentinel-2's spectral capabilities, this invention not only monitors salinization in its static state but also tracks its evolution as a dynamic process. This perspective allows tracking the evolution of soil salinization over time, detecting early signs of degradation, and providing actionable insights for land management in the Hetao Irrigation District.

[0097] In this paper, the 2021 Sentinel-1 / 2 collaborative data was used to invert saline-alkali soils in cultivated land. The spectral data obtained by Sentinel-2 is shown in Table 2.

[0098] Table 2 Optical remote sensing data from Sentinel-2

[0099] Band name describe Resolution Central wavelength B2 Blue 10m 496.6nm(S2A) / 492.1nm(S2B) B3 Green 10m 560nm(S2A) / 559nm(S2B) B4 Red 10m 664.5nm(S2A) / 665nm(S2B) B5 RedEdge1 20m 703.9nm(S2A) / 703.8nm(S2B) B6 RedEdge2 20m 740.2nm(S2A) / 739.1nm(S2B) B7 RedEdge3 20m 782.5nm(S2A) / 779.7nm(S2B) B8 Near-infrared (NIR) 10m 835.1nm(S2A) / 833nm(S2B) B8A RedEdge4 20m 864.8nm(S2A) / 864nm(S2B) B11 SWIR1 20m 1613.7nm(S2A) / 1610.4nm(S2B) B12 SWIR2 20m 2202.4nm(S2A) / 2185.7nm(S2B)

[0100] The present invention carried out field sampling operations on the soil in the Hetao Irrigation District in 2021. The specific sampling methods and quantities are as follows:

[0101] Sampling point layout: The sampling scope selected plots larger than 100 mu in the study area to ensure the spatial representativeness of the plots and cover the main soil salinization types in the area; in each plot, sampling points were arranged with a sampling interval greater than 10 meters to avoid spatial autocorrelation between sampling points and ensure the independence of sampling points; 0-10 cm surface soil was sampled at each sampling point, and a mixed sampling method was used to reduce sampling errors; three locations were selected around the sampling point (usually distributed in an equilateral triangle or a ring), and soil samples were collected separately; the soil samples at the three locations were mixed evenly to form a mixed sample.

[0102] Sampling sample processing: The weight of each mixed sample is about 1 kg, ensuring sufficient sample volume for subsequent determination; after the soil samples are collected, they are placed in clean plastic bags, and the sampling number, coordinates (latitude and longitude), plot number, sampling time and other information of each sample are recorded in detail.

[0103] Laboratory determination: After eliminating invalid soil samples, a total of 487 soil samples were determined, and the distribution is as follows: Figure 2 Electrical conductivity (EC) measurements are performed by a third-party professional testing agency. Method: Each soil sample is mixed with deionized water at a fixed ratio (e.g., a 1:5 soil-to-water ratio), allowed to stand, and the supernatant extracted. The conductivity of the soil supernatant is measured using a professional conductivity meter. EC alone is measured as a key indicator for analyzing soil salinity. Once the results are returned, they are combined with the sampling point location information for subsequent analysis and model validation.

[0104] Data preprocessing: Cloud detection and masking: Clouds and cirrus are removed using the quality assurance band (QA60) of Sentinel-2. Multi-temporal image synthesis: Images within the selected time range are averaged and synthesized to reduce data noise. Radar data processing: A 3×3 Boxcar filter is applied to the radar data to smooth noise.

[0105] After obtaining multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data in the Hetao Irrigation District, the remote sensing feature combination was optimized using the recursive feature elimination method based on the random forest algorithm and the radar and optical remote sensing data. The model performance of the optimized remote sensing feature combination was evaluated using cross-validation to obtain the optimal remote sensing feature combination, which is as follows:

[0106] The REF algorithm is used to perform recursive feature elimination and cross-validation, such as Figure 3 As shown in Figure 2, the changes in the main accuracy indicator RMSE under different feature combinations are shown. It can be seen that when setting 5, 10, 15, 20, and all feature combinations, the RMSE is only 0.938 when the 5-feature combination is used, reaching the highest prediction level. In other words, the REF algorithm extracts the best-performing feature combination for subsequent modeling based on the random forest algorithm. The optimal feature combination after recursive verification is Greenness, Wetness, VV, B11, and NDVI, as shown in Figure 2. Figure 4 shown.

[0107] On the premise of obtaining the optimal feature set supported by REF-CV, the sample data set was split into 0.8, that is, 80% as the training set and 20% as the validation set. Based on the optimal feature set (Greenness, Wetness, VV, B11 and NDVI), regression prediction was performed on the validation data set (20% of the samples) based on the random forest algorithm. Figure 4 This figure shows the results of a feature variable importance analysis, commonly used to assess the relative importance of input variables in a random forest model. %IncMSE represents the relative impact of each variable on the model's prediction error, typically measured by removing the variable and observing the change in mean squared error (MSE). The horizontal axis represents %IncMSE, which is the percentage increase in model prediction error after removing the variable. A larger value indicates a greater impact on the model's prediction performance, and therefore a higher importance. As can be seen from the figure, Wetness has the greatest impact on model prediction error, making it the most important variable. VV comes in second. NDVI and Greenness are of similar importance. B11 has the least impact on model prediction error. IncNodePurity indicates the degree to which each variable improves node purity and is commonly used to measure a variable's importance during the decision tree splitting process. A higher node purity indicates that the variable is more effective at distinguishing data. The horizontal axis represents IncNodePurity, with larger values indicating a greater contribution to model building. As can be seen from the figure, Greenness has the greatest impact on node purity, making it the most important variable. VV and Wetness contribute less. NDVI and B11 are relatively less important.

[0108] Overall, wetness and greenness are the two most important variables in the model, dominating the error impact %IncMSE and node purity improvement IncNodePurity, respectively. VV radar vertical polarization (strong) is also an important variable, ranking second. NDVI and B11 are less important, but still contribute to model building. Furthermore, the validation results of the random forest regression prediction on the 20% sample are as follows: Figure 5 shown.

[0109] Through hyperparameter adjustment, the optimal random forest regression model configuration was finally obtained, as shown in Table 3.

[0110] Table 3 Optimal random forest regression model hyperparameters

[0111]

[0112] Based on the random forest regression model with the hyperparameter adjustment in Table 3, a regression prediction of the salinization degree of cultivated land soil in the Hetao Irrigation District in 2021 was conducted. The details are as follows:

[0113] In this study, a comparative analysis of different band and feature combinations ultimately identified an optimal combination of Greenness, Wetness, VV, B11, and NDVI. This combination comprehensively considers the spectral response characteristics of crop growth and salinization, as well as the sensitivity of vegetation indices. This helps improve the model's ability to accurately predict soil salinization, particularly in the complex context of salinization dynamics in arid and semi-arid regions.

[0114] During the model building phase (2021), the model was rigorously trained and validated using training samples, ultimately achieving an average R 2 The inversion results of soil salinization degree of cultivated land in Hetao Irrigation District in 2021 are as follows: Figure 6 shown.

[0115] Example 3:

[0116] Embodiment 3 of the present invention discloses a soil salinity inversion system based on multimodal remote sensing data fusion using a soil salinity inversion method based on multimodal remote sensing data fusion, comprising:

[0117] Data acquisition module: used to obtain radar remote sensing data and optical remote sensing data of the area where soil salinity is to be inverted, and to measure the electrical conductivity of the soil as an inversion indicator of soil salinity;

[0118] Feature screening module: used to obtain multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data, and to optimize the remote sensing feature combination based on the random forest algorithm and the recursive feature elimination method based on the combination of radar remote sensing data and optical remote sensing data. The model performance of the optimized remote sensing feature combination is evaluated by cross-validation to obtain the optimal remote sensing feature combination.

[0119] Salinity inversion module: used to construct and train a soil conductivity prediction model based on the random forest algorithm based on the optimal remote sensing feature combination, input the test data into the soil conductivity prediction model after hyperparameter adjustment, obtain soil conductivity, and complete soil salinity inversion.

[0120] Example 4:

[0121] Embodiment 4 of the present invention discloses an electronic device, including:

[0122] memory for storing computer programs;

[0123] The processor is configured to implement the steps of a soil salinity inversion method based on multimodal remote sensing data fusion when executing the computer program.

[0124] Example 5:

[0125] Embodiment 5 of the present invention discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the computer program implements the steps of a soil salinity inversion method based on multimodal remote sensing data fusion.

[0126] The embodiments of the present invention disclose a soil salinity inversion method, system, device and storage medium based on multimodal remote sensing data fusion. The present invention extracts multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data by integrating two complementary Sentinel-1 radar data (providing insights into soil moisture and surface characteristics) and Sentinel-2 optical data (capturing vegetation and soil spectral characteristics related to salinity), and optimizes the feature combination and evaluates the performance through recursive feature elimination and cross-validation methods. The optimal feature combination (Greenness, Wetness, VV, B11 and NDVI) suitable for the most typical Hetao Irrigation District is obtained. Finally, the hyperparameter-adjusted model is applied to the soil salinity inversion in the Hetao Irrigation District in 2021, proving the accuracy of the present invention. In summary, the present invention achieves a more accurate inversion of soil salinity under the influence of complex environments and the interaction of human factors.

[0127] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0128] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A soil salinity inversion method based on multimodal remote sensing data fusion, characterized in that: include: Step 1: Obtain radar remote sensing data and optical remote sensing data of the area where soil salinity is to be inverted, and measure the soil electrical conductivity as an inversion indicator of soil salinity; Step 2: Obtain multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data. Combine the radar remote sensing data with the optical remote sensing data, and optimize the remote sensing feature combination using the recursive feature elimination method based on the random forest algorithm. Use cross-validation to evaluate the model performance of the optimized remote sensing feature combination to obtain the optimal remote sensing feature combination. Step 3: Based on the optimal remote sensing feature combination, a soil conductivity prediction model based on the random forest algorithm is constructed and trained. The measured data is input into the soil conductivity prediction model after hyperparameter adjustment to obtain soil conductivity and complete soil salinity inversion.

2. The soil salinity inversion method based on multimodal remote sensing data fusion according to claim 1, characterized in that: In step 1, the radar remote sensing data includes: VV band and VH band.

3. The soil salinity inversion method based on multimodal remote sensing data fusion according to claim 1, characterized in that: In step 1, the optical remote sensing data includes: blue band B2, green band B3, red band B4, first red edge band B5, second red edge band B6, third red edge band B7, near infrared band B8, fourth red edge band B8A, first shortwave infrared band B11 and second shortwave infrared band B12.

4. The soil salinity inversion method based on multimodal remote sensing data fusion according to claim 1, characterized in that: In step 2, the multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data include: normalized difference vegetation index NDVI, enhanced vegetation index EVI, soil adjusted vegetation index SAVI, modified soil adjusted vegetation index MSAVI, slope adjusted vegetation index TSAVI, adjusted normalized vegetation index OSAVI, adaptive soil adjusted vegetation index ATSAVI, vegetation index PVI, near infrared vegetation index NIRv, chlorophyll index MTCI, water sensitive vegetation index SSAI, chromaticity index Clr, vegetation health index CIg, first normalized difference index NDRE1, second normalized difference index NDRE2, first normalized difference vegetation index NDVIr1, second normalized difference vegetation index NDVIr2, third normalized difference vegetation index NDVIr3, radar vegetation index VV / VH Ratio, radar vegetation index RVI, dual-polarization SAR vegetation index DPSVI, normalized difference polarimetric index NDPI, a spectral transformation method Brightness, greenness index Greenness and wetness index Wetness.

5. The soil salinity inversion method based on multimodal remote sensing data fusion according to claim 1, characterized in that: In step 2, based on the random forest algorithm, the recursive feature elimination method is used to optimize the remote sensing feature combination, specifically: In each iteration, the importance score of the feature is calculated by using the Gini index, and the feature with the least contribution to the model is eliminated; The Gini index is an indicator for measuring impurity. For a node t, the Gini index is as follows: Where C is the total number of categories; p i is the proportion of samples belonging to category i to the total samples of the node; The smaller the Gini index, the purer the samples in the node, which means that most samples belong to the same category; When a feature is used to split a node, the data is divided into two child nodes, left and right. The Gini gain measures the reduction in the Gini index after the split, as follows: Among them, Gini(t) is the Gini index of the node before splitting; Gini(t left ) and Gini(t right ) are the Gini indexes of the left and right child nodes after splitting; N left and N right are the number of samples of the left and right child nodes after splitting; N is the total number of samples of the current node; The larger the Gini gain, the greater the contribution of the feature to the classification; In a random forest, each split node of each tree calculates the Gini gain based on a certain feature. The importance score of the feature is the cumulative average of the Gini gains of all trees in the forest, as follows: Where T is the total number of trees in the random forest; is the Gini gain of feature j in the k-th tree in one split; Random Forest evaluates the importance of each feature using the above formula and performs recursive feature elimination.

6. The soil salinity inversion method based on multimodal remote sensing data fusion according to claim 1, characterized in that: In step 2, cross-validation is used to evaluate the model performance of the optimized remote sensing feature combination, specifically: The 5-fold cross-validation method was used to randomly divide the sample data into 5 mutually exclusive subsets. Four of the subsets were used as training sets each time, and the remaining subset was used as the test set for a total of 5 iterations. In each iteration, the random forest model is used to train the training set, and the trained model is used to predict the test set to obtain the predicted value and the true value. The average determination coefficient and root mean square error of 5 iterations are used to evaluate the model performance of different remote sensing feature combinations.

7. The soil salinity inversion method based on multimodal remote sensing data fusion according to claim 1, characterized in that: In step 3, the hyperparameter adjustment includes: the number of trees in the random forest ntrees, the maximum number of nodes per tree maxnodes, the minimum number of samples of the terminal node nodesize, and the number of samples extracted from the training set for each tree sampsize.

8. A soil salinity inversion system based on multimodal remote sensing data fusion using the soil salinity inversion method based on multimodal remote sensing data fusion according to any one of claims 1 to 7, characterized in that: include: Data acquisition module: used to obtain radar remote sensing data and optical remote sensing data of the area where soil salinity is to be inverted, and to measure the electrical conductivity of the soil as an inversion indicator of soil salinity; Feature screening module: used to obtain multimodal remote sensing features based on the fusion of radar remote sensing data and optical remote sensing data, and to optimize the remote sensing feature combination based on the random forest algorithm and the recursive feature elimination method based on the combination of radar remote sensing data and optical remote sensing data. The model performance of the optimized remote sensing feature combination is evaluated by cross-validation to obtain the optimal remote sensing feature combination. Salinity inversion module: used to construct and train a soil conductivity prediction model based on the random forest algorithm based on the optimal remote sensing feature combination, input the test data into the soil conductivity prediction model after hyperparameter adjustment, obtain soil conductivity, and complete soil salinity inversion.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor is configured to implement the steps of a soil salinity inversion method based on multimodal remote sensing data fusion as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the soil salinity inversion method based on multimodal remote sensing data fusion as described in any one of claims 1 to 7 are implemented.

Images