Soil salinity mapping method for quantitatively introducing vegetation type based on single thermal code

Abstract

The invention discloses a soil salinity mapping method for quantitatively introducing a vegetation type based on single thermal coding. The method comprises the following steps: acquiring a remote sensing image, field actually measured soil salinity and corresponding vegetation type data; spectrum parameters sensitive to salinity are extracted and screened; processing the vegetation type by adopting single thermal coding, and converting the vegetation type into an unbiased binary feature vector; the spectrum parameters and the vegetation feature vectors are fused to serve as input, and an inversion model is constructed and trained; applying the model to whole-region data to generate a soil salinity spatial distribution map; according to the method, a pseudo-sequence relation introduced by traditional coding is eliminated through single thermal coding, so that the model can learn the internal relation between vegetation and salinity in an unbiased manner, and the precision and reliability of soil salinity inversion in a complex earth surface environment are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of soil remote sensing detection technology, and in particular to a method for soil salinity mapping based on single-thermal coding and quantitative introduction of vegetation types. Background Technology

[0002] Soil salinization is a critical ecological problem threatening global land resources and sustainable agricultural development, especially in coastal and delta regions with complex topographic and hydrological conditions, where the need for dynamic evolution monitoring and precise management is particularly urgent. Therefore, obtaining large-scale, high-precision spatial distribution information of soil salinity is a prerequisite for effective management of salinization. Currently, quantitative remote sensing technology, with its macroscopic, rapid, and non-destructive characteristics, has replaced the time-consuming and labor-intensive traditional field sampling and chemical analysis methods, becoming the mainstream technology for regional-scale soil salinity monitoring. Early remote sensing inversion methods mainly relied on the direct relationship between surface spectral information and soil salinity content to build models. However, because the salinity spectral response is easily affected by the coupling of various factors such as land cover and soil moisture, the model accuracy and stability are limited. To improve inversion accuracy, subsequent studies have begun to introduce quantifiable environmental variables, such as groundwater depth, topographic factors, and meteorological data, as auxiliary information. Multi-source data fusion has enhanced the model's interpretability to some extent.

[0003] However, existing technologies still face core technical bottlenecks when dealing with qualitative environmental variables that significantly influence soil salinity distribution, particularly vegetation type. Some studies have attempted to quantify the categorical variable of vegetation type and introduce it into the model using traditional coding techniques such as ordinal coding and mean coding. These coding methods require manually setting numerical values ​​or orders, such as simply assigning different vegetation types values ​​like 1, 2, 3, etc. This erroneously introduces magnitude relationships and numerical distances between originally independent categories, i.e., "spurious distances." This artificially introduced pseudo-order does not match the true physical relationship between categories and salinity, inputting incorrect prior information into the model. This interferes with the model's accurate learning of the complex nonlinear relationship between "vegetation type-soil salinity," thus limiting the accuracy and generalization ability of the inversion model and making it difficult to meet the needs of high-precision salinity mapping in complex surface environments.

[0004] Therefore, how to scientifically and unbiasedly quantify and integrate key qualitative factors such as vegetation type is a technical problem that urgently needs to be solved to overcome the current bottleneck in the accuracy of soil salinity remote sensing inversion. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0006] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a soil salinity mapping method based on monothermic coding and quantitative introduction of vegetation types to solve the problems mentioned in the background art.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a soil salinity mapping method based on single-thermal coding and quantitative introduction of vegetation type, comprising: Acquire remote sensing image data of the study area, field measured soil data including geographic coordinates and soil salinity, and vegetation type data corresponding to the geographic coordinates; Based on the remote sensing image data, spectral parameters are extracted, and by analyzing the correlation between the spectral parameters and the soil salinity content, sensitive spectral parameters are screened out. The vegetation types in the study area are then subjected to single-thermal encoding, and each vegetation type is converted into an independent binary vector to generate a vegetation type feature vector. The sensitive spectral parameters and the vegetation type feature vector are fused to form the input features of the inversion model. The input features and the field measured soil data are used to construct and train an inversion model for predicting soil salinity. The inversion model is applied to remote sensing image data and vegetation type data of the study area to generate a spatial distribution map of soil salinity in the study area.

[0008] As a preferred embodiment of the soil salinity mapping method based on single-thermal coding quantitative introduction of vegetation type described in this invention, the field measured soil data includes soil salinity content data at at least two different soil depths.

[0009] As a preferred embodiment of the soil salinity mapping method based on single-thermal coding quantitative introduction of vegetation type described in this invention, wherein: 1. The process of screening sensitive spectral parameters includes: Perform correlation analysis and retain spectral parameters whose absolute values ​​of correlation coefficients with soil salinity content are greater than the preset correlation threshold; Additionally, perform multicollinearity analysis to remove parameters from the retained spectral parameters whose autocorrelation coefficients are greater than a preset redundancy threshold.

[0010] As a preferred embodiment of the soil salinity mapping method based on single-thermal coding quantitative introduction of vegetation type described in this invention, wherein: the single-thermal coding processing method is as follows: If there are N vegetation types in the study area, each vegetation type is converted into a binary vector of length N, where the element at the index position corresponding to the vegetation type itself in the binary vector has a value of 1, and the element at the other N-1 positions has a value of 0.

[0011] As a preferred embodiment of the soil salinity mapping method based on single-thermal coding quantitative introduction of vegetation type described in this invention, wherein: the sensitive spectral parameter and the vegetation type feature vector are fused in the following manner: The values ​​of the sensitive spectral parameters are concatenated with the vegetation type feature vector to form a combined feature vector, which is used as the input feature.

[0012] As a preferred embodiment of the soil salinity mapping method based on single-thermal coding quantitative introduction of vegetation type described in this invention, the inversion model is selected from at least one of the following: multiple linear regression model, BP neural network model, random forest model, and support vector machine model.

[0013] As a preferred embodiment of the soil salinity mapping method based on single-thermal coding quantitative introduction of vegetation type described in this invention, wherein: when the inversion model is a BP neural network model, the model is trained by setting its learning rate, maximum number of iterations and target error threshold.

[0014] As a preferred embodiment of the soil salinity mapping method based on single-heat coding quantitative introduction of vegetation type described in this invention, wherein: when the inversion model is a support vector machine model, a radial basis function is used as its kernel function, and the model is trained by setting a penalty factor, a gamma parameter, and an insensitive loss function parameter.

[0015] As a preferred embodiment of the soil salinity mapping method based on single-thermal coding and quantitative introduction of vegetation type described in this invention, the method further includes: constructing and training an inversion model for predicting soil salinity content using the input features and the field-measured soil data; and further comprising: Multiple candidate inversion models are constructed, and the performance of the multiple candidate inversion models is evaluated using at least one of the evaluation indicators of the coefficient of determination, root mean square error, relative prediction bias and linear correlation coefficient. The optimal inversion model is selected from these models and used as the inversion model for remote sensing image data and vegetation type data in the study area.

[0016] As a preferred embodiment of the soil salinity mapping method based on single-thermal coding quantitative introduction of vegetation type described in this invention, wherein: for the soil salinity content at the first soil layer depth, a BP neural network model is used as the optimal inversion model; for the soil salinity content at the second soil layer depth, which is greater than the first soil layer depth, a support vector machine model is used as the optimal inversion model.

[0017] Compared with existing technologies, the beneficial effects of this solution are: 1. To address the problem that existing technologies, when processing qualitative environmental variables such as vegetation type, introduce "spurious distances" through sequential encoding, which interfere with model learning and thus limit inversion accuracy, this invention uses single-hot encoding to process vegetation types. By converting each vegetation type into mutually independent binary vectors with equal distances, artificially introduced pseudo-order relationships and numerical biases are completely eliminated. This allows the machine learning model to learn the true intrinsic relationship between spectral features and soil salinity under different vegetation backgrounds without bias, thereby significantly improving the prediction accuracy and generalization ability of the soil salinity inversion model. 2. By collecting soil samples at different depths and performing inversion tasks for different soil salinity contents, the machine learning model that best matches the physical causes was selected, realizing refined and differentiated modeling of the vertical distribution of soil salinity. This allows the final salinity distribution map to not only reflect the surface salinity status but also reveal the salinity distribution pattern in the subsurface, providing richer information dimensions and offering more in-depth scientific basis for agricultural irrigation management and three-dimensional prevention and control of salinization. 3. Furthermore, this invention also selects spectral parameters that are most sensitive to salt response and have the lowest information redundancy by combining correlation analysis with multicollinearity diagnosis (VIF), effectively eliminating noise and redundant information. The selected spectral parameters are then spliced ​​and fused with unbiased quantized vegetation type feature vectors, ensuring the feature quality of the input model, reducing the complexity of model training, enhancing the stability and interpretability of the model, and guaranteeing the reliability of the final mapping results. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the overall process of a soil salinity mapping method based on single-thermal coding and quantitative introduction of vegetation types, as described in one embodiment of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0023] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0024] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0025] Example 1 Reference Figure 1This is the first embodiment of the present invention, which provides a soil salinity mapping method based on single-thermal coding and quantitative introduction of vegetation types, including: S1. Acquire remote sensing image data of the study area, field measured soil data including geographic coordinates and soil salinity, and vegetation type data corresponding to geographic coordinates. It should be noted that this step is the basic data preparation stage for building a high-precision soil salinity inversion model. The purpose is to obtain multi-source heterogeneous data that can reflect the soil salinity status and its influencing factors. The quality and representativeness of this multi-source heterogeneous data directly determine the upper limit of model training and the reliability of the final mapping results. Furthermore, multispectral remote sensing images that cover the entire study area, have good imaging quality, and low cloud cover are selected as the primary data source. Examples include imagery from the ZY1 series satellites (ZY1-02C) or the Gaofen-7 satellite. These imagery data possess moderate spatial resolution and rich spectral bands, effectively capturing the spectral reflectance characteristics of the Earth's surface. Simultaneously, to ensure the accuracy of spectral information, a series of preprocessing steps are required on the original images before extracting spectral parameters, including: Radiometric calibration: converting dimensionless count values ​​from raw image records into apparent radiance of the top layer of the atmosphere with actual physical meaning; Atmospheric correction: Eliminate or reduce the absorption and scattering effects of atmospheric molecules and aerosols on electromagnetic waves, and invert the apparent radiance of the top atmospheric layer into the true reflectance of the surface. Since the surface reflectance is a physical quantity that truly reflects the spectral characteristics of matter, it is the physical basis for constructing a quantitative inversion model. Mature correction models such as FLAASH can be used to accomplish this. Geometric correction and cropping: Orthorectification and georegistration are performed on the image to ensure that the position of each pixel on the image corresponds precisely to the actual geographic coordinates. The corrected image is then cropped according to the administrative or natural boundaries of the study area to reduce the amount of data in subsequent processing. Furthermore, the field-measured soil data serves as the "ground truth" for model construction and validation. Within the study area, soil sampling points were established following the principles of scientific rigor, representativeness, and uniformity. At each sampling point, a handheld GPS device was used to accurately record its latitude and longitude coordinates. To comprehensively reflect the vertical distribution characteristics of soil salinity, stratified soil sampling was necessary. Specifically, at each sampling point, soil samples were collected at two depths: the first soil layer (e.g., topsoil, 0–20 cm) and the second soil layer (e.g., subsurface soil, 20–40 cm), both deeper than the first layer. The collected soil samples were sealed, preserved, and brought back to the laboratory. After air-drying, grinding, and sieving, the conductivity (EC) of the 1:5 soil-to-water ratio extract was measured using the conductivity method. It should be noted that conductivity is a key indicator characterizing the total amount of soluble salt ions in the soil and is highly positively correlated with soil salinity. Subsequently, based on a regional empirical conversion model, the conductivity value was converted to total dissolved salt (TDS). Salts (TSS) content, usually expressed in g / kg, is used as the soil salinity content for model training in this invention. Specifically, to establish the conversion relationship between electrical conductivity (EC) and total dissolved salts (TSS) content, a subset (e.g., 30 samples) can be randomly selected from all collected soil samples, and two measurements can be performed simultaneously: first, the electrical conductivity (EC) of the 1:5 soil-to-water ratio extract is measured using the aforementioned method; second, the total dissolved salts (TSS) content is determined by gravimetric method (drying and evaporation method). Subsequently, using the measured EC value as the independent variable and the TSS value as the dependent variable, regression analysis (e.g., univariate linear regression) is performed to establish a conversion model specific to this study area, such as... .in, and As regression coefficients, this transformation model can then be used to quickly convert EC measurements of all other samples into TSS content, serving as the true values ​​of soil salinity for model training. Furthermore, since vegetation is a key environmental factor influencing surface spectral response and soil water and salt transport, and its type has a complex ecological relationship with soil salinity, the vegetation type covering the surface was simultaneously investigated and recorded at each sampling point while soil sampling was being conducted. For example, in coastal wetland environments, vegetation types may include reeds, tamarisk, Suaeda salsa, crops, or bare land. Then, the recorded vegetation types were associated with the corresponding geographic coordinates of the sampling points to form a comprehensive ground-measured dataset of "coordinates-salinity content-vegetation type". This process adopted a simultaneous field survey method to ensure a high degree of spatial consistency between vegetation type data and soil salinity data, providing high-quality input for the subsequent unbiased integration of this qualitative information into the inversion model through single-thermal coding. S2. Extract spectral parameters based on remote sensing image data, and screen out sensitive spectral parameters by analyzing the correlation between spectral parameters and soil salinity content. Perform single-thermal coding on vegetation types in the study area, convert each vegetation type into an independent binary vector, and generate vegetation type feature vectors. It should be noted that the purpose of this step is to extract and construct input variables from the raw data that can efficiently and accurately characterize the soil salinity status. The step includes two parallel processing branches: one is the quantitative spectral feature extraction and optimization for remote sensing images, and the other is the unbiased quantification of qualitative environmental variables (vegetation type). Furthermore, the extraction process of spectral parameters is as follows: Multidimensional spectral parameters are extracted from preprocessed remote sensing images to form a comprehensive initial feature set that reflects multiple information about the Earth's surface. These spectral parameters mainly include three categories: Raw band reflectance: Directly extract the surface reflectance values ​​of each spectral band (such as blue, green, red, near-infrared, short-wave infrared, etc.) in the image; Vegetation Indices: To quantify the shading effect of surface vegetation cover on the soil spectrum, at least one vegetation index is constructed and extracted. For example, the Normalized Difference Vegetation Index (NDVI) is calculated by dividing the difference between near-infrared reflectance and red reflectance by the sum of the two. Its expression is as follows: in, This represents the surface reflectance in the near-infrared band. The NDVI value represents the surface reflectance in the red light band. The higher the NDVI value, the more lush the vegetation growth and the higher the coverage, which helps the model distinguish the attenuation of salinity information caused by vegetation cover. Salinity indices: To enhance the sensitivity to soil salinity information, various existing salinity indices that have been proven to have a certain indicative role in salinity are constructed and extracted, such as Salinity Index I (SI-I) and Normalized Difference Salinity Index (NDSI). These indices, through combination calculations of specific bands, aim to amplify the weak spectral features of salt crusts, salt deposits, or high-salinity soils, thereby improving the signal-to-noise ratio. Specifically, the formula for calculating the salinity index I (SI-I) is as follows: in, Indicates the surface reflectance in the green light band; Furthermore, since the initial feature set extracted above has a high dimensionality, it may contain parameters that are not strongly correlated with soil salinity (noise) or have highly overlapping information between features (redundancy). In order to improve the training efficiency, stability and generalization ability of the model, feature screening is required. The process of screening out sensitive spectral parameters is as follows: Correlation analysis was performed: The Pearson correlation coefficient was used to measure the linear correlation between each spectral parameter and the field-measured soil salinity obtained in step S1. The calculation expression is as follows: in, The correlation coefficient, The total number of sampling points. For the first The value of a certain spectral parameter corresponding to each sampling point This represents the measured soil salinity at this location. and Let the average values ​​of the spectral parameter and soil salinity at all sampling points be given, and calculate the values ​​of all spectral parameters. After setting the value, a preset correlation threshold is set, with the threshold range between (0,1), for example, 0.3, to retain only the absolute value of the correlation coefficient. Spectral parameters exceeding this threshold are initially eliminated for features that are unrelated to or weakly correlated with salt content. Perform multicollinearity analysis: For parameters that passed the correlation analysis, further diagnose multicollinearity issues by using the variance inflation factor (VIF). The VIF calculation formula is as follows: in, It is the first The variance expansion factor of each spectral parameter It is the first The coefficient of determination is obtained when a spectral parameter is used as the dependent variable and all other retained spectral parameters are used as independent variables in a linear regression. The larger the VIF value, the stronger the collinearity between the parameter and other parameters. A preset redundancy threshold is set, which is usually between 5 and 20, for example, 10. From the retained spectral parameters, parameters with VIF values ​​greater than the threshold are iteratively removed until the VIF values ​​of all remaining parameters are less than the threshold. This eliminates information redundancy between features, ensures that the spectral parameters input into the model are relatively independent, and makes the model weights more interpretable, avoiding model instability caused by collinearity. It should be noted that after the above two steps of screening, the final result is the set of sensitive spectral parameters that have the most significant response to soil salinity and the lowest information redundancy. Furthermore, to address the problem of "spurious distance" introduced by traditional encoding methods, which limits model performance, the present invention employs one-hot encoding to convert discrete, unordered categorical variables into numerical vectors that are easily processed by the model, while maintaining the independence between categories. The process of one-hot encoding vegetation types is as follows: The total number of vegetation types involved in all sampling points obtained in step S1 is counted, denoted as N. Then, a unique index position (from 1 to N) is created for each vegetation type. For any vegetation type at a sampling point, it is converted into a binary vector of length N. In this vector, the element at the index position corresponding to the vegetation type itself has a value of 1, while the elements at the other N-1 positions have values ​​of 0. For example, if there are 5 main types in the study area: reeds, tamarisk, salt flat lichen, crops, and bare land (N=5), then: If a sampling point is "reed", its vegetation type feature vector is: [1, 0, 0, 0, 0]; If a sampling point is "Tamarix", its vegetation type feature vector is: [0, 1, 0, 0, 0]; If a sampling point is “Suaeda salsa”, its vegetation type feature vector is: [0, 0, 1, 0, 0]; If a sampling point is a "crop", its vegetation type feature vector is: [0, 0, 0, 1, 0]; If a sampling point is "bare land", its vegetation type feature vector is: [0, 0, 0, 0, 1]; It should be noted that in this way, each vegetation type is mapped to a point in a high-dimensional space, and the Euclidean distance and cosine similarity between any two vectors of different types are fixed, which perfectly solves the "spurious distance" problem. This allows the machine learning model to learn the unique relationship between each vegetation type and soil salinity as an independent, unordered feature, without being interfered with by incorrect human sorting, thus laying the foundation for building a more accurate inversion model. The final generated binary vector is the vegetation type feature vector. S3. The sensitive spectral parameters and vegetation type feature vectors are fused to form the input features of the inversion model. Using the input features and field measured soil data, an inversion model for predicting soil salinity is constructed and trained. It should be noted that this step aims to accurately establish a nonlinear mapping relationship from remotely observable features to soil salinity content; Furthermore, the feature fusion process is as follows: The set of sensitive spectral parameters selected in step S2 is fused with the generated vegetation type feature vector to form a unified combined feature vector containing multi-dimensional information, which serves as the sole input to the subsequent inversion model. Specifically, the fusion is performed as feature splicing. Assuming that after the screening in step S2, M sensitive spectral parameters are obtained (such as... Given that there are N vegetation types within the study area, for any sampling point, its input feature vector... The composition is as follows: in, It is a vector composed of the specific values ​​of the M sensitive spectral parameters corresponding to that point; It is a single-hot encoded vector of the vegetation type corresponding to the point, with a length of N, and only one element is 1, while the rest are 0; It should be noted that by using this splicing method, continuous spectral information and discrete category information are placed in the same feature space, enabling the model to not only perceive "what kind of light is reflected by the ground surface" and "under what vegetation cover this reflection behavior occurs" during the learning process, but also to learn the differential relationship between spectral parameters and soil salinity under different vegetation backgrounds, thereby effectively overcoming the model bias caused by relying solely on spectral information or incorrect encoding. Furthermore, the process of constructing and training the inversion model is as follows: All the ground-based measured datasets of "coordinates-salt content-vegetation type" obtained in step S1 are randomly divided into training and validation sets according to a certain ratio (e.g., 70% as training set and 30% as validation set). It should be explained that the training set is used for model parameter learning and construction, and the validation set is used for model performance evaluation and hyperparameter tuning to test the model's generalization ability. Based on this training set, a variety of candidate inversion models are constructed. These candidate inversion models can cover different types from simple to complex, such as: multiple linear regression model (MLR), BP neural network model (BPNN), random forest model (RF) and support vector machine model (SVM). The purpose of constructing a variety of candidate inversion models is to find the inversion model that best matches the input features constructed by the present invention and has the best performance through comparison. Training and performance evaluation were conducted for each candidate model, with BP neural network and support vector machine as examples: When the inversion model is a BP neural network, this model performs nonlinear fitting by simulating the structure of neurons in the human brain. Its training process continuously adjusts the network weights and biases through the backpropagation algorithm to minimize the error between the predicted and true values. Key training parameters of this model include: learning rate, used to control the step size of weight updates in each iteration; maximum number of iterations, used to prevent excessive training time or getting trapped in local optima; and target error threshold, serving as the stopping condition for training convergence. By adjusting these parameters, the BP neural network model can fully learn the complex relationship between input features and soil salinity. Nonlinear relationships; when the inversion model is a support vector machine model, this model is based on statistical learning theory and is particularly good at handling small sample, high-dimensional and nonlinear problems. In this embodiment, the radial basis function (RBF) is preferred as its kernel function to map the original features to a higher-dimensional space, thereby solving the linearly inseparable problem; its training process mainly finds the optimal classification hyperplane by optimizing the following hyperparameters: penalty factor, used to balance the complexity of the model and the training error, and control the tolerance for misclassified samples; gamma parameter, used to define the influence range of a single training sample; and insensitive loss function parameter, used to define the width of the "pipeline" in the regression model where the loss is not calculated; Specifically, in this embodiment, the learning rate of the BP neural network model is preferably set between 0.01 and 0.1, for example, 0.05; the maximum number of iterations is preferably set between 500 and 2000, for example, 1000; and the target error threshold is preferably set to less than 10⁻⁻⁶. 4 For example, 10⁻ 5 ; Specifically, in this embodiment, for the support vector machine model, methods such as cross-validation and grid search are used to determine the optimal parameter combination. The preferred search range for the penalty factor is [0.1, 100], the preferred search range for the gamma parameter is [0.01, 10], and the preferred search range for the insensitive loss function parameter is [0.01, 1]. For example, in this embodiment, a preferred parameter combination is: penalty factor = 10, gamma parameter = 0.1, and insensitive loss function parameter = 0.01. The performance of all trained candidate inversion models is evaluated using a validation set, and the optimal inversion model is selected. This performance evaluation employs quantitative metrics, including but not limited to: Coefficient of determination ( The value is used to measure how well the model explains the variance of the data. The closer the value is to 1, the better the model fit. Its expression is: in, To determine the number of samples in the validation set, For the model to the first The predicted value for each sample, This is its measured value. The average of all measured values; Root Mean Square Error (RMSE): Measures the magnitude of the deviation between the predicted and actual values. The smaller the value, the higher the model accuracy. Its expression is: Relative prediction bias (RPD): the ratio of the standard deviation of the measured value to the predicted RMSE. The larger the RPD, the stronger the model's predictive ability. Linear correlation coefficient: used to measure the linear correlation between predicted and measured values; Specifically, in a preferred embodiment of the present invention, the optimal model selection for different soil depths is obtained through the above evaluation process: For the inversion of soil salinity at the first soil layer depth (e.g., 0-20cm), the BP neural network model shows the best performance because the surface soil salinity is subject to intense and complex interactions from various surface factors such as vegetation, evaporation, and irrigation, and the relationship between them is highly nonlinear. The powerful nonlinear fitting ability of the BP neural network can precisely capture this complexity. For the inversion of soil salinity at the second soil layer depth (e.g., 20-40 cm), the support vector machine model becomes the optimal choice because the changes in deep soil salinity are relatively gradual and less affected by instantaneous surface dynamics. Although its relationship with remotely observable features (including long-term water and salt environment indicated by vegetation type) is still nonlinear, it is more structural. SVM, based on the principle of minimizing structural risk, has better robustness and generalization ability when dealing with such problems, and thus can more stably reveal the intrinsic connection between deep salinity and comprehensive surface features. It should be noted that through the above systematic model construction, training, evaluation and optimization, the optimal inversion model that can most accurately predict soil salinity content after introducing single-thermal coding vegetation type features was finally determined, laying the foundation for generating a high-precision salinity distribution map of the whole region. S4. Apply the inversion model to the remote sensing image data and vegetation type data of the study area to generate a spatial distribution map of soil salinity in the study area. It should be noted that this step aims to generalize the optimal inversion model trained in step S3, which is based on point samples, to the continuous space of the entire study area, thereby achieving a leap from discrete sampling points to a full-coverage, high-precision spatial distribution map. Furthermore, in order for the model to make predictions for every pixel within the study area, the input features for each pixel must first be prepared to be exactly the same as those used during model training. This includes: Calculation of sensitive spectral parameters for the whole area: For the remote sensing image that has been preprocessed and covers the entire study area, which was acquired in step S1, all sensitive spectral parameters selected in step S2 are calculated pixel by pixel. For example, if NDVI and SI-I are selected as sensitive spectral parameters, then the NDVI value and SI-I value of each pixel in the image need to be calculated to generate a sensitive spectral parameter raster layer covering the whole area. Generation of vegetation type feature vectors for the entire region: This requires spatial distribution data of vegetation types covering the entire study area, i.e., a vegetation classification map. This map can use the vegetation type data with geographic coordinates obtained from the field survey in step S1 as training and validation samples. The remote sensing image data covering the entire region (e.g., ZY1-02C multispectral image) is processed using a supervised classification algorithm (such as a support vector machine classifier or a random forest classifier) ​​to generate a vegetation classification map of the entire region with the same spatial resolution as the remote sensing image. In this map, each pixel is assigned a vegetation type (e.g., reed, tamarisk, bare land, etc.). Then, for each pixel in the vegetation classification map, according to its assigned vegetation type, the one-hot encoding rule in step S2 is applied to generate its corresponding vegetation type feature vector. For example, a pixel classified as "tamarisk" has a feature vector of [0, 1, 0, 0, 0] (assuming there are 5 vegetation types and the tamarisk index is 2). Fusion of input features across the entire area: Following the feature fusion method in step S3, for each pixel in the study area, the calculated sensitive spectral parameter value is concatenated with the generated vegetation type feature vector. This results in a multi-dimensional combined feature vector matrix covering the entire study area. Each row of this matrix represents a pixel, and each column represents an input feature dimension, thus preparing the data for the model's pixel-by-pixel prediction. Furthermore, the optimal inversion model trained and selected in step S3, such as a BP neural network model for the first soil layer depth or a support vector machine model for the second soil layer depth, is loaded into the computing environment. The combined feature vector matrix generated in the previous step, covering the entire study area, is used as the input of the model to predict the soil salinity content of each pixel in the study area. The output of the model is a raster data file with the same spatial resolution and georegistration information as the original remote sensing image, where the value of each pixel is the predicted soil salinity content (TSS, unit g / kg) at that geographic location. Furthermore, since the salinity prediction raster data obtained in the previous step is quantitative, but for intuitive presentation and practical application of the results, it needs to be converted into a visualized thematic map, including: Salinization classification: Based on relevant industry standards (such as soil environmental quality standards) or specific research needs, different salt content thresholds are set to divide continuous salt prediction values ​​into several discrete salinization levels, such as: non-salinization, slight salinization, moderate salinization, severe salinization, and saline soil. Map symbolization and embellishment: Assign clearly distinguishable colors to different salinity levels (e.g., use a gradient color spectrum from green to red) to visually express the spatial differentiation of salinity content, and add map embellishment elements such as titles, legends (explaining the salinity level and salinity range represented by each color), scale bars, and north arrows in Geographic Information System (GIS) software. It should be noted that, after the above processing, a spatial distribution map of soil salinity in the study area that is rich in information and easy to read can be generated. This map can transform complex models and data into a high-precision spatial information product that can provide key decision-making basis for regional land resource planning, optimization of agricultural production layout, and precise management of salinization.

[0026] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0027] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0028] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0029] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0030] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0031] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A soil salinity mapping method based on single-thermal encoding quantification of vegetation types, characterized by, The application relates to a method for predicting soil salt content based on remote sensing image data and vegetation type data. The method comprises the following steps: acquiring remote sensing image data of a research area, field measured soil data containing geographical coordinates and soil salt content, and vegetation type data corresponding to the geographical coordinates; extracting spectral parameters based on the remote sensing image data, screening sensitive spectral parameters by analyzing the correlation between the spectral parameters and the soil salt content, performing single-hot encoding processing on the vegetation types in the research area, converting each vegetation type into an independent binary vector, and generating a vegetation type feature vector; fusing the sensitive spectral parameters and the vegetation type feature vector to form input features of an inversion model, and using the input features and the field measured soil data to build and train the inversion model for predicting soil salt content; 2. The soil salinity mapping method based on single-heat encoding quantification of vegetation types according to claim 1, wherein, applying the inversion model to the remote sensing image data and the vegetation type data of the research area to generate a soil salt spatial distribution map of the research area.

3. The method for mapping soil salinity incorporating vegetation type based on single-thermal encoding quantification according to claim 1, wherein, The field measured soil data comprises soil salt content data of at least two different soil layer depths. The process of screening the sensitive spectral parameters comprises the following steps: performing correlation analysis to retain spectral parameters with an absolute value of a correlation coefficient greater than a preset correlation threshold value; 4. The method for mapping soil salinity incorporating vegetation type based on single-thermal encoding quantification according to claim 1, wherein, and performing multiple collinearity analysis to eliminate parameters with an autocorrelation coefficient greater than a preset redundancy threshold value from the retained spectral parameters. The single-hot encoding processing mode is as follows:

5. The method for soil salinity mapping based on single-thermal encoding quantification of introduced vegetation types according to claim 1, wherein, if there are N vegetation types in the research area, each vegetation type is converted into a binary vector with a length of N, wherein the element value of the index position corresponding to the vegetation type itself in the binary vector is 1, and the element values of the other N-1 positions are 0. The fusion mode of the sensitive spectral parameters and the vegetation type feature vector is as follows:

6. The soil salinity mapping method based on single-heat coding quantification of introduction of vegetation types according to claim 1 or 5, characterized by, the numerical values of the sensitive spectral parameters are spliced with the vegetation type feature vector to form a combined feature vector as the input features.

7. The method for mapping soil salinity incorporating vegetation type based on single-heat encoding quantification according to claim 6, wherein, The inversion model is selected from at least one of a multiple linear regression model, a BP neural network model, a random forest model and a support vector machine model.

8. The method for mapping soil salinity incorporating vegetation type based on single-heat encoding quantification according to claim 6, wherein, When the inversion model is a BP neural network model, the model is trained by setting a learning rate, a maximum iteration number and a target error threshold value.

9. The method for soil salinity mapping based on single-heat encoding quantification of introduced vegetation types according to claim 1, wherein, When the inversion model is a support vector machine model, a radial basis function is used as a kernel function of the model, and the model is trained by setting a penalty factor, a gamma parameter and an insensitive loss function parameter. Using the input features and the field measured soil data to build and train the inversion model for predicting soil salt content further comprises the following steps: building multiple candidate inversion models, performing performance evaluation on the multiple candidate inversion models by using at least one evaluation index of a determination coefficient, a root mean square error, a relative prediction deviation and a linear correlation coefficient, optimally selecting an optimal inversion model from the multiple candidate inversion models, and using the optimal inversion model as the inversion model applied to the remote sensing image data and the vegetation type data of the research area.

10. The soil salinity mapping method based on single-heat coding quantification of introduction of vegetation types according to claim 2 or 9, characterized by, The BP neural network model is used as the optimal inversion model for the soil salt content of the first soil layer depth; and the support vector machine model is used as the optimal inversion model for the soil salt content of the second soil layer depth greater than the first soil layer depth.

Images