Soil organic matter prediction mapping method, device and equipment based on deep learning and medium

Abstract

The invention discloses a soil organic matter prediction mapping method, device and equipment based on deep learning and a medium, and relates to the field of digital soil mapping, and the method comprises the steps: obtaining a remote sensing image of target region soil, and setting a preset number of soil sampling points; comprehensively acquiring environmental data indexes such as terrain factors, vegetation factors, climate factors and land utilization factors; constructing three-dimensional environment tensor data in a preset range of each sampling point, and inputting the three-dimensional environment tensor data into the soil organic matter content prediction model to obtain an organic matter content prediction value set of the soil sampling points in the target area; and generating a spatial distribution prediction map of the soil organic matter content of the target area by combining the prediction value set and the spatial coordinates of the soil sampling points, thereby improving the accuracy of soil organic matter prediction and the mapping efficiency, and providing technical support for soil resource management and agricultural sustainable development.

Description

Technical Field

[0001] This application relates to the field of digital soil mapping, and particularly to a method, device, equipment and medium for predicting and mapping soil organic matter based on deep learning. Background Art

[0002] With the acceleration of the global urbanization process, the role of urban agriculture in maintaining urban ecological balance, ensuring food safety, and promoting sustainable economic development has become increasingly prominent. The development of urban agriculture requires accurate knowledge of the soil fertility information of cultivated land to optimize land use efficiency, rationally allocate agricultural resources, and formulate precise agricultural management measures.

[0003] Soil fertility attributes, especially soil organic matter content, are key indicators for measuring soil quality. Their spatial distribution characteristics have a direct impact on crop yields and soil health conditions, which is particularly important for small-scale cultivated land with high density distribution in cities. Urban cultivated land often shows high fragmentation and heterogeneity, and the spatial variation of soil attributes is affected by a variety of factors.

[0004] Related soil fertility prediction methods, such as interpolation methods and linear regression models, have limited accuracy when dealing with complex and variable environmental conditions. Although machine learning algorithms such as random forests have a certain ability to model non-linear relationships, their utilization of spatial context information is insufficient, and the prediction effect depends on the quality of covariate selection. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, equipment and medium for predicting and mapping soil organic matter based on deep learning, which can effectively capture the non-linear relationship between soil organic matter content and factor index data, and achieve high-precision prediction of soil organic matter content and rapid drawing of spatial distribution maps.

[0006] To achieve the above purpose, this application provides the following solutions:

[0007] In the first aspect, this application provides a method for predicting and mapping soil organic matter based on deep learning, including:

[0008] Obtain remote sensing images of the soil within the target area, and set a preset number of soil sampling points within the target area;

[0009] Obtain the environmental data indicators of the soil within the target area; the environmental data indicators include topographic factors, vegetation factors, climate factors, and land use factors; the topographic factors are calculated through a digital elevation model and include slope aspect, topographic wetness index, and curvature; the vegetation factors are obtained through normalization based on remote sensing image data and include the normalized difference vegetation index; the climate factors are obtained based on remote sensing images and meteorological data and include land surface temperature, temperature-vegetation drought index, and Palmer drought severity index; the land use factors are further calculated based on the interpreted data of remote sensing images through the land use type map obtained by classification processing and include paddy fields, irrigated land, and dry land;

[0010] Extract the environmental data indicators within the preset range of each soil sampling point respectively to obtain a plurality of three-dimensional environmental tensor data; each three-dimensional environmental tensor data includes the environmental data indicator vectors of all pixels within the preset range; the environmental data indicator vector includes the numerical values of all environmental data indicators, and the length of the environmental data indicator vector is equal to the number of environmental data indicators;

[0011] Input the plurality of three-dimensional environmental tensor data into the soil organic matter content prediction model in sequence to obtain a set of predicted values of the soil organic matter content of the soil sampling points within the target area; the soil organic matter content prediction model is obtained by iteratively training a preset deep learning network model using a sample data set; the sample data set includes the historical three-dimensional environmental tensor data of historical soil sampling points and the corresponding measured values of the soil organic matter content;

[0012] According to the set of predicted values of the soil organic matter content of the soil sampling points within the target area and the spatial coordinate values of the soil sampling points on the remote sensing image, obtain the spatial distribution prediction map of the soil organic matter content within the target area.

[0013] In a second aspect, the present application provides a soil organic matter prediction mapping device based on deep learning, including:

[0014] A remote sensing image and soil sampling point acquisition module, configured to acquire the remote sensing image of the soil within the target area and set a preset number of soil sampling points within the target area;

[0015] An environmental data index acquisition module for acquiring environmental data indexes of soil within the target area; the environmental data indexes include topographic factors, vegetation factors, climate factors, and land use factors; the topographic factors are calculated through a digital elevation model and include slope aspect, topographic wetness index, and curvature; the vegetation factors are obtained through normalization based on remote sensing image data and include the normalized difference vegetation index; the climate factors are obtained based on remote sensing images and meteorological data and include land surface temperature, temperature vegetation drought index, and Palmer drought severity index; the land use factors are further calculated based on the interpreted data of remote sensing images through a classified land use type map and include paddy fields, irrigated land, and dry land;

[0016] An environmental data index extraction module for respectively extracting environmental data indexes within a preset range of each soil sampling point to obtain a plurality of three-dimensional environmental tensor data; each three-dimensional environmental tensor data includes an environmental data index vector of all pixels within the preset range; the environmental data index vector includes the numerical values of all environmental data indexes, and the length of the environmental data index vector is equal to the number of environmental data indexes;

[0017] A soil organic matter content prediction module for sequentially inputting the plurality of three-dimensional environmental tensor data into a soil organic matter content prediction model to obtain a set of predicted values of the soil organic matter content of soil sampling points within the target area; the soil organic matter content prediction model is obtained by iteratively training a preset deep learning network model using a sample data set; the sample data set includes historical three-dimensional environmental tensor data of historical soil sampling points and corresponding measured values of the soil organic matter content;

[0018] A spatial distribution prediction map generation module for obtaining a spatial distribution prediction map of the soil organic matter content within the target area according to the set of predicted values of the soil organic matter content of soil sampling points within the target area and the spatial coordinate values of the soil sampling points on the remote sensing image.

[0019] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the steps of the method for predicting and mapping soil organic matter based on deep learning as described in any one of the above.

[0020] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the method for predicting and mapping soil organic matter based on deep learning as described in any one of the above.

[0021] According to the specific embodiments provided by the present application, the present application has the following technical effects:

[0022] The present application provides a method, apparatus, device, medium and product for predicting and mapping soil organic matter based on deep learning. By obtaining remote sensing images of soil within the target area and determining soil sampling points and extracting environmental data indicators for each sampling point, the problem of incomplete and inaccurate data acquisition is solved. Using remote sensing technology and geographic information systems, combined with multi-source data such as terrain, vegetation, climate and land use, a comprehensive and accurate environmental data indicator system is constructed. By extracting environmental data indicators for each sampling point and constructing three-dimensional environmental tensor data, the problems of how to extract effective information from massive data and how to associate this information with soil organic matter content are solved, providing reliable data support for the training of deep learning models. By inputting the three-dimensional environmental tensor data into the soil organic matter content prediction model, high-precision prediction of soil organic matter content is achieved. The prediction model is constructed based on a deep learning network and can learn the complex relationship between environmental data and soil organic matter content through iterative training of the sample data set. By combining the spatial coordinate values of the sampling points, the predicted soil organic matter content is presented in the form of a spatial distribution map, making the results more intuitive and easy to understand. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0024] Figure 1 It is a schematic flowchart of a method for predicting and mapping soil organic matter based on deep learning provided by an embodiment of the present application;

[0025] Figure 2 It is a structural diagram of a soil organic matter content prediction model provided by an embodiment of the present application;

[0026] Figure 3 It is a regression fitting diagram of the predicted value and the measured value of the organic matter content in the test set of the soil organic matter content prediction model provided by an embodiment of the present application;

[0027] Figure 4 It is a schematic diagram of the functional modules of a device for predicting and mapping soil organic matter based on deep learning provided by an embodiment of the present application.

[0028] Figure 5 It is a schematic structural diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] First, some technical terms involved in the embodiments of the present application are introduced.

[0030] Soil organic matter: A comprehensive organic substance that includes plant and animal residues, microorganisms, and their metabolites in the soil, and can play an important role in providing nutrients, retaining moisture, and stabilizing soil structure.

[0031] Digital Soil Mapping (DSM): A technical process based on the soil-landscape relationship, which combines means such as Geographic Information System (GIS), remote sensing, and statistics, aims to predict the spatial distribution characteristics of soil properties, and generates a soil spatial variation map in a rasterized form, and is an important tool for modern soil mapping research.

[0032] Prediction accuracy: An index that measures the closeness between the calculation results of a digital mapping model and the true value, usually characterized by statistical quantities such as Coefficient of Determination (R 2 ), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). The higher the prediction accuracy, the more reliable the model.

[0033] Kriging interpolation method: A spatial interpolation method based on geostatistics, which analyzes the spatial autocorrelation between sample points through a variogram, and optimally predicts the soil property values of unsampled points in a way that minimizes the estimation error. It mainly includes methods such as ordinary Kriging, simple Kriging, and co-Kriging interpolation.

[0034] Inverse distance weighted method: A deterministic interpolation method that assumes that the influence of neighboring sample points on the prediction point is inversely proportional to the distance, and performs spatial prediction by assigning different weights to the sample points.

[0035] Random Forest Regression (RF): An ensemble learning algorithm that constructs multiple decision trees and combines multiple input variables to predict continuous target values. It is suitable for dealing with the non-linear relationship of soil properties.

[0036] Convolutional Neural Networks (CNN): A deep learning model that is good at extracting spatial context information from images or raster data, and can capture complex spatial heterogeneity and multi-modal covariate features in soil mapping.

[0037] Remote sensing: A technology that uses sensors to obtain electromagnetic wave information of the Earth's surface in a non-contact manner from a distance. By detecting the reflection and radiation signals of the target object, it is widely used to monitor environmental variables such as soil, vegetation, and water bodies.

[0038] NDVI (Normalized Difference Vegetation Index): A vegetation index calculated from the reflectance of the red and near-infrared bands, used to evaluate vegetation coverage and health status.

[0039] DEM (Digital Elevation Model): A digital dataset representing the height distribution of the Earth's surface, usually provided in raster form, which can be used to calculate topographic parameters (such as slope and aspect) and applied to soil mapping.

[0040] Curvature: A shape feature of the terrain surface, aiming to describe the bending degree and change trend of the terrain slope at a certain point.

[0041] TWI (Topographic Wetness Index): An index used to characterize the ability of the terrain to accumulate water. It combines the information of catchment area and slope, and is used to describe the potential of a certain terrain unit to accumulate water and the distribution of soil moisture.

[0042] LST (Land Surface Temperature): Temperature information obtained through remote sensing or field measurement, which reflects the thermal environment status of the Earth's surface and has an important impact on soil moisture evaporation and plant growth.

[0043] TVDI (Temperature Vegetation Dryness Index): A drought monitoring index based on temperature and vegetation index, commonly used to evaluate soil moisture status and drought degree.

[0044] PDSI (Palmer Drought Severity Index): A meteorological drought assessment index, which calculates the severity of regional drought based on soil moisture balance.

[0045] Digital soil mapping is a research field at the intersection of soil science and geographic information technology. By modeling the data of sampling points and environmental covariates, it predicts the spatial distribution of soil properties. Compared with traditional manual soil mapping, DSM has advantages such as high efficiency, accuracy, and flexibility, and is applicable to various spatial scales, from small farmlands to large regional areas. Currently, the mainstream mapping methods applied in DSM include two major categories: spatial interpolation and regression algorithms. Among them, spatial interpolation methods include ordinary Kriging, co-Kriging, and inverse distance weighting, etc., which are applicable to areas with many sampling points and a small range. Regression algorithms mainly include traditional linear regression and machine learning-based nonlinear regression. Traditional linear regression has high requirements for the independence assumption of covariates and is difficult to adapt to the changes in soil properties in complex environments. Therefore, currently, algorithms such as random forest, support vector machine, and gradient boosting decision tree are mainly used in regression algorithms. Machine learning methods can better adapt to complex terrains or heterogeneous regions by modeling the relationship between covariates and soil properties. The random forest regression algorithm is currently the most widely used in soil mapping by machine learning, but it also has certain limitations, that is, it is difficult to capture spatial context information and is relatively dependent on the selection of covariates.

[0046] The current mainstream mapping methods of digital soil mapping (DSM) include two major categories: spatial interpolation and regression algorithms (such as Kriging interpolation, random forest regression, etc.). Although these methods have achieved certain results in predicting the spatial distribution of soil properties, when facing complex urban environments and highly fragmented cultivated lands, the following technical problems still exist:

[0047] 1. Insufficient handling of spatial heterogeneity: Urban cultivated lands are usually highly fragmented, and soil properties are strongly affected by various factors, showing great spatial heterogeneity. Traditional interpolation methods and machine learning models are difficult to effectively capture this complexity.

[0048] 2. Limited ability to model nonlinear relationships: There are usually nonlinear and multivariate interaction relationships between soil properties (such as organic matter) and environmental covariates. Some interpolation methods (such as inverse distance weighting, etc.) and linear regression methods (such as multiple linear regression, support vector regression, etc.) lack sufficient nonlinear modeling ability.

[0049] 3. Insufficient utilization of spatial context information: The soil properties of urban cultivated lands are affected by comprehensive factors such as topographic features, vegetation cover, climate conditions, and human activities, while both interpolation methods and machine learning lack the direct processing ability for rasterized spatial data (such as remote sensing images).

[0050] 4. Limited ability to integrate multi-modal data: Most algorithms can only process numerical or raster data, and have limited ability to integrate various modal information such as remote sensing data (NDVI, TVDI), terrain data (DEM, curvature, aspect), and human activity data (land use type, etc.) required for urban cultivated land soil prediction.

[0051] Based on the above analysis, the main bottleneck of current digital soil mapping in urban complex environments lies in how to improve the accuracy and adaptability of soil property mapping in highly heterogeneous and fragmented areas. Specifically, the following technical problems need to be solved: First, how to capture the non-linear relationship between soil properties and multi-modal covariates in complex urban environments. This problem requires breaking through the limitations of traditional interpolation and regression methods and using more advanced models to model the complex associations of multi-source data. Second, how to improve mapping accuracy by using spatial context information. The deficiencies of current methods in spatial feature extraction make it difficult to handle complex spatial distribution patterns in urban environments. Therefore, there is an urgent need for a method that can combine rasterized data (such as remote sensing images) to make full use of spatial context information. Third, how to achieve efficient fusion of multi-modal data. For different types of data (numerical, raster, etc.), a unified framework needs to be developed to effectively integrate the data and improve mapping accuracy.

[0052] Deep learning is a subfield of machine learning. In recent years, deep learning has been preliminarily applied to the field of geodetic surveying and mapping, but it is still in the initial research stage in the field of digital soil mapping. Considering the ability of convolutional neural networks (CNNs) in deep learning to extract spatial features of images and raster data, applying it to soil mapping has great potential. CNNs can extract high-order features of input data through multi-layer convolutional operations, incorporating the spatial context information around soil sampling points into the model, so as to more accurately capture complex soil-environment relationships. The soil mapping method based on deep learning provides a new idea for solving the high-precision mapping requirements in urban complex environments, and also provides stronger technical support for cultivated land management and agricultural production.

[0053] In view of this, the embodiments of the present application provide a soil organic matter prediction mapping method based on deep learning, which uses a convolutional neural network (CNN) to achieve high-precision prediction of soil properties in the complex environment of urban cultivated land, providing a new idea for solving the high-precision mapping requirements in urban complex environments, and also providing stronger technical support for cultivated land management and agricultural production.

[0054] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0055] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0056] In an exemplary embodiment, as Figure 1 shown, a method for predicting and mapping soil organic matter based on deep learning is provided, including the following steps 101 to 105. Among them:

[0057] Step 101, obtain remote sensing images of the soil within the target area range, and set a preset number of soil sampling points within the target area range. First, determine the target mapping area range according to research requirements, and divide the area boundary through a geographic information system (GIS) tool or remote sensing images. Obtain remote sensing image data related to the target area, preferably remote sensing images with a spatial resolution of 30 meters or higher (such as Landsat 8 images), and obtain multi-temporal images throughout the year according to the research period to ensure data integrity and timeliness. Then, use the grid point method or stratified random sampling method to arrange soil sampling points according to the mapping area range and expected sampling density. The spacing of the soil sampling points can be adjusted according to the size and spatial heterogeneity of the research area. Accurately locate each soil sampling point, and it is recommended to use a GPS device to record its longitude and latitude coordinates.

[0058] Step 102, obtain environmental data indicators of the soil within the target area range; the environmental data indicators include topographic factors, vegetation factors, climate factors, and land use factors; the topographic factors are calculated through a digital elevation model and include aspect, topographic wetness index, and curvature; the vegetation factors are obtained through normalization based on remote sensing image data and include the normalized difference vegetation index; the climate factors are obtained based on remote sensing images and meteorological data and include surface temperature, temperature-vegetation drought index, and Palmer drought severity index; the land use factors are further calculated based on the interpreted data of remote sensing images through the land use type map obtained by classification processing and include paddy fields, irrigated land, and dry land. It may also include soil factors, indicating various soil type variables such as paddy soil, fluvo-aquic soil, and yellow brown soil. The specific calculation method of the environmental data indicators is selected according to the variable characteristics.

[0059] Step 103: Extract the environmental data indicators within the preset range of each soil sampling point respectively to obtain multiple three-dimensional environmental tensor data; each three-dimensional environmental tensor data includes the environmental data indicator vectors of all pixels within the preset range; the environmental data indicator vector includes the numerical values of all environmental data indicators, and the length of the environmental data indicator vector is equal to the number of environmental data indicators. Use GIS software (such as ArcGIS) to process the environmental factors of the target area, and uniformly resample them to the same spatial resolution as the remote sensing image (such as 30 meters). Extract the environmental factor data within a certain range (such as 15×15 pixels) around each sampling point to form a three-dimensional tensor structure (spatial dimension × variable dimension).

[0060] Step 104: Input the multiple three-dimensional environmental tensor data into the soil organic matter content prediction model in sequence to obtain the set of predicted values of the soil organic matter content of the soil sampling points within the target area; the soil organic matter content prediction model is obtained by iteratively training a preset deep learning network model using a sample data set; the sample data set includes the historical three-dimensional environmental tensor data of historical soil sampling points and the corresponding measured values of the soil organic matter content. The input layer of the soil organic matter content prediction model is the normalized three-dimensional environmental tensor data. Use 2-3 convolutional modules, combined with the ReLU activation function and the max pooling operation, to extract the spatial context features, and then perform dimensionality reduction processing on the output of the convolutional layer through a fully connected layer. Finally, complete the regression prediction of the soil organic matter content through the output layer.

[0061] The training set and validation set of the soil organic matter content prediction model can randomly divide the soil sampling point data set according to 80% and 20%. Evaluate the model performance through indicators such as the coefficient of determination (R 2 ) and the root mean square error (RMSE), and select the best model through multiple trainings.

[0062] Step 105: According to the set of predicted values of the soil organic matter content of the soil sampling points within the target area and the spatial coordinate values of the soil sampling points on the remote sensing image, obtain the spatial distribution prediction map of the soil organic matter content within the target area. Use the trained soil organic matter content prediction model (CNN model) and the normalized environmental data indicators to perform pixel-by-pixel prediction of the required soil organic matter content over the entire area, and superimpose the prediction results on the spatial coordinates to generate the spatial distribution prediction map of the soil organic matter content within the target area.

[0063] By implementing the above steps 101 to 105, this application can significantly improve the accuracy and efficiency of soil organic matter prediction and generate a high-precision spatial distribution prediction map of the soil organic matter content.

[0064] In another exemplary embodiment of the present application, before the above step 103, the method may further include: normalizing each three-dimensional environmental tensor data using the min-max normalization method. All environmental factor data are normalized to the interval [0, 1] using the min-max normalization method.

[0065] As an alternative implementation, the soil organic matter content prediction model in step 104 includes: an input layer, three convolutional pooling layers, two fully connected layers, and an output layer connected in sequence.

[0066] The input layer is used to receive three-dimensional environmental tensor data.

[0067] The convolutional pooling layer is used to extract local features from the input three-dimensional environmental tensor data through convolution operations and reduce the dimension of the three-dimensional environmental tensor data through pooling operations.

[0068] The fully connected layer is used to perform dimensionality reduction processing on the output of the convolutional pooling layer.

[0069] The output layer is used to calculate the predicted value of the organic matter content of each soil sampling point within the target area range according to the output of the fully connected layer using a regression algorithm.

[0070] When implementing this implementation, the method for obtaining the measured value of the soil organic matter content in step 104 specifically includes:

[0071] At each historical soil sampling point, an initial soil sample at a depth of 0-20 cm is collected by the plum blossom-shaped multi-point mixing method.

[0072] Perform preprocessing operations on the initial soil sample to obtain a preprocessed soil sample; the preprocessing operations include air-drying and grinding and sieving the initial soil sample to remove the mixed bricks, tiles, stones, and animal and plant residues in the soil sample.

[0073] Use the potassium dichromate oxidation-volumetric method to measure the measured value of the soil organic matter content in the preprocessed soil sample.

[0074] As an alternative implementation, the Kjeldahl method can also be used to measure the measured value of total soil nitrogen. Other physical or chemical properties are detected according to actual needs.

[0075] In another exemplary embodiment of the present application, the calculation formulas for the slope aspect, terrain wetness index, and curvature in the terrain factor are:

[0076]

[0077] Among them, Aspect represents the slope aspect of the raster cell; z is the elevation, representing the height of the raster cell; represents the elevation change rate of the slope surface of the grid cell in the x direction, where x represents the east-west direction; represents the elevation change rate of the slope surface of the grid cell in the y direction, where y represents the north-south direction; TWI represents the terrain wetness index of the grid cell; α represents the catchment area of the grid cell; tanβ represents the tangent value of the slope of the grid cell; Curvature represents the curvature of the grid cell; is the second derivative of the slope of the grid cell in the x direction, representing the degree of bending of the terrain of the grid cell in the x direction; is the second derivative of the slope of the grid cell in the y direction, representing the degree of bending of the terrain of the grid cell in the y direction.

[0078] As an optional implementation manner, the calculation formula of the normalized difference vegetation index in the vegetation factor is:

[0079]

[0080] where NDVI represents the normalized difference vegetation index of the grid cell; NIR represents the reflectance of the near-infrared band of the grid cell; RED represents the reflectance of the red band of the grid cell.

[0081] As an optional implementation manner, the calculation formulas of the land surface temperature, temperature vegetation drought index, and Palmer drought severity index in the climate factor are respectively:

[0082]

[0083] where LST represents the land surface temperature of the grid cell; λ represents the effective wavelength of the thermal infrared band of the grid cell; ρ represents the constant of the grid cell; ε represents the land surface emissivity of the grid cell; T B represents the brightness temperature of the grid cell; K 1 represents the first band radiation constant of the grid cell, with a value of 1; K 2 represents the second band radiation constant of the grid cell, with a value of 2; L λ represents the radiance value of the grid cell.

[0084]

[0085] where PDSI represents the temperature vegetation drought index of the grid cell; d represents the actual water supply-demand deviation of the grid cell; PE represents the potential evapotranspiration of the grid cell; R represents the surface runoff of the grid cell; L represents the soil moisture deficit of the grid cell; α 1 β, and γ respectively represent the first empirical coefficient, the second empirical coefficient, and the third empirical coefficient, which are used to adjust the contributions of different water parameters to the humidity index; k represents the climate adaptation coefficient.

[0086]

[0087] Among them, TVDI represents the Palmer Drought Severity Index of the grid cell; LST min is the minimum land surface temperature, representing the wet boundary of the grid cell; LST max is the maximum land surface temperature, representing the dry boundary of the grid cell.

[0088] In another exemplary embodiment of this application, taking a certain city as an example, the application process of this solution is actually demonstrated.

[0089] (1) Soil sample collection and determination of organic matter content.

[0090] In November 2021, using the cultivated land distribution map of a certain city obtained by remote sensing interpretation as the base map, 94 effective soil sampling points were arranged and collected by the grid sampling method, and GPS was used for navigation and positioning to the soil sampling points. The sampling depth of the samples was 0 - 20 cm, and the multi-point mixed sampling method in a plum blossom shape was used for sampling. One kilogram of samples was taken by the quartering method, and the sampling time was November 2021. After the collected soil samples were air-dried indoors, the mixed animal and plant residues in the soil were picked out, ground and sieved for testing. The actual measured value of soil organic matter content was determined by the potassium dichromate oxidation - volumetric method.

[0091] (2) Extraction and screening of environmental factors.

[0092] Based on data such as remote sensing images, terrain, and soil type maps, environmental factors associated with soil organic matter were obtained.

[0093] ① Remote sensing image data: Landsat 8 data with a 30-meter spatial resolution was selected as the remote sensing data source, and the acquisition time was the whole year of 2021. The terrain data came from the global 30-meter resolution digital elevation model data of SRTM.

[0094] ② Screening and acquisition of environmental factors: Environmental data indicators affecting soil organic matter content were extracted from four aspects: terrain, vegetation, climate, and human activities. After performing a correlation analysis on all the extracted environmental data indicators and soil organic matter content, the factors with too low correlation were excluded. Finally, 10 environmental data indicators of environmental factors were screened out, as shown in Table 1. Among them, the terrain factors include aspect, terrain humidity, and curvature, which were extracted from the DEM; the vegetation factors were calculated using remote sensing data, and the climate factors were calculated through the surface dynamic feedback variables calculated by remote sensing; the land use type factors include three cultivated land types: paddy fields, irrigated land, and dry land, which were obtained from the land use remote sensing interpretation data of a certain city in 2021.

[0095] Table 1 Description and source of environmental factors

[0096]

[0097] The calculation formulas for each environmental factor are as follows:

[0098]

[0099] Among them, Aspect represents the slope aspect of the grid cell; z is the elevation, representing the height of the grid cell; represents the elevation change rate of the slope surface of the grid cell in the x direction, where x represents the east-west direction; represents the elevation change rate of the slope surface of the grid cell in the y direction, where y represents the north-south direction; TWI represents the topographic wetness index of the grid cell; α represents the catchment area of the grid cell; tanβ represents the tangent value of the slope of the grid cell; Curvature represents the curvature of the grid cell; is the second derivative of the slope of the grid cell in the x direction, representing the degree of bending of the terrain of the grid cell in the x direction; is the second derivative of the slope of the grid cell in the y direction, representing the degree of bending of the terrain of the grid cell in the y direction.

[0100] As an optional implementation method, the calculation formula for the normalized difference vegetation index in the vegetation factor is:

[0101]

[0102] Among them, NDVI represents the normalized difference vegetation index of the grid cell; NIR represents the reflectance of the near-infrared band of the grid cell; RED represents the reflectance of the red band of the grid cell.

[0103] As an optional implementation method, the calculation formulas for land surface temperature, temperature vegetation drought index, and Palmer drought severity index in the climate factor are respectively:

[0104]

[0105] Among them, LST represents the land surface temperature of the grid cell; λ represents the effective wavelength of the thermal infrared band of the grid cell; ρ represents the constant of the grid cell; ε represents the land surface emissivity of the grid cell; T B represents the brightness temperature of the grid cell; K 1 represents the first band radiation constant of the grid cell, with a value of 1; K 2 represents the second band radiation constant of the grid cell, with a value of 2; L λ represents the radiance value of the grid cell.

[0106]

[0107] Among them, PDSI represents the temperature-vegetation drought index of the grid cell; d represents the actual water supply-demand deviation of the grid cell; PE represents the potential evapotranspiration of the grid cell; R represents the surface runoff of the grid cell; L represents the soil moisture deficit of the grid cell; α 1 , β, and γ represent the first empirical coefficient, the second empirical coefficient, and the third empirical coefficient respectively, which are used to adjust the contributions of different moisture parameters to the humidity index; k represents the climate adaptation coefficient.

[0108]

[0109] Among them, TVDI represents the Palmer drought severity index of the grid cell; LST min is the minimum surface temperature, representing the wet boundary of the grid cell; LST max is the maximum surface temperature, representing the dry boundary of the grid cell.

[0110] Paddy field, dummy variable. 1 indicates that the land use type of the grid cell is paddy field; 0 indicates that the land use type of the grid cell is not paddy field.

[0111] Irrigated land, dummy variable. 1 indicates that the land use type of the grid cell is irrigated land; 0 indicates that the land use type of the grid cell is not irrigated land.

[0112] Dry land, dummy variable. 1 indicates that the land use type of the grid cell is dry land; 0 indicates that the land use type of the grid cell is not dry land.

[0113] ③ Normalization of environmental factors: Use ArcGIS 10.8 software to resample all environmental factors to the size of a certain city at a resolution of 30m, generating a total of 10 tif files. Use Matlab software to read all the data and convert it to the [0, 1] interval through the min-max normalization method for the convenience of CNN model training.

[0114] (3) Soil organic matter content prediction model based on convolutional neural network (CNN).

[0115] The CNN model is used to capture the complex non-linear relationship between soil organic matter and multi-modal environmental factors and its spatial context information. This application constructs and trains a CNN model based on soil sampling point data and environmental data indicators.

[0116] ① Model input design: The input structure of the CNN is a spatial neighborhood tensor containing 15×15 pixels (the neighborhood size is 15×15), and each pixel contains the values of 10 environmental factors (see the description of environmental factors for details). Among them, the environmental factors that can be calculated include Aspect, TWI, Curvature, NDVI, LST, PDSI, and TDVI, and the environmental factors obtained through dummy variable transformation include three land use types: paddy fields, irrigated land, and dry land.

[0117] Construction of three-dimensional environmental tensor data: Extract the values of environmental factors in the 15×15 neighborhood around each sampling point to construct the three-dimensional environmental tensor data (15×15×10) of the input data.

[0118] ② CNN network structure, as Figure 2 shown. Input layer: Receive the 15×15×10 input three-dimensional environmental tensor data.

[0119] Convolution and pooling module: It includes 3 convolutional blocks, and each convolutional block consists of a convolutional layer, a ReLU activation function, and a max pooling layer, which are used to extract features.

[0120] Fully connected layer: Through two fully connected layers, the high-dimensional features output by the convolutional layer and the pooling layer are dimensionally reduced and mapped to the sample label space.

[0121] Output layer: Used for regression prediction, and the output is the normalized value of soil organic matter.

[0122] ③ Model training and evaluation: Data division: 94 soil sampling points are divided into a training set and a test set, and are randomly divided according to a ratio of 0.8:0.2, which are used for model training and verification respectively.

[0123] Optimization method: The Adam optimizer is adopted, the initial learning rate is 0.0001, and the maximum number of training epochs is 200.

[0124] Evaluation index: The performance of the model is evaluated by the coefficient of determination (R 2 ), root mean square error (RMSE), and mean absolute error (MAE). The specific calculation formulas are as follows:

[0125]

[0126] Among them, y i is the measured value of soil organic matter content at the i-th sample point; is the predicted value of soil organic matter content at the i-th sample point; is the average value of the measured values of soil organic matter content.

[0127] ④ Spatial prediction distribution: Input rasterized environmental data indicators (such as NDVI, TVDI, terrain humidity index, etc.), process them in blocks and input them into the trained CNN model for prediction.

[0128] Spatial overlay: Combine the model prediction results with the spatial coordinates of the sampling points to generate a prediction map of the spatial distribution of the complete soil organic matter content.

[0129] ⑤ Model prediction results and evaluation: The regression fitting graph of the predicted value and the measured value of the organic matter content in the CNN model test set is as Figure 3 shown. The abscissa represents the measured value of the soil organic matter content, and the ordinate represents the predicted value of the soil organic matter content (obtained from model prediction). The dotted line is the image of y = x. When the coordinates fall near the dotted line, it represents better fitting performance.

[0130] This application compares the CNN model with ordinary Kriging interpolation, inverse distance weighting, and random forest models. The comparison results are shown in Table 2.

[0131] Table 2 Comparison results of the accuracy of four models

[0132] Model category <![CDATA[R 2 > RMSE MAE Ordinary Kriging 0.50 7.19 5.78 Inverse Distance Weighting 0.51 6.67 5.34 Random Forest Model 0.60 4.35 3.87 CNN Model 0.64 4.21 3.79

[0133] According to the results in Table 2, the prediction accuracy of the CNN model is better than that of the other three types of models. Its determination coefficient R 2 is 0.61, the root mean square error RMSE is 4.51, and the mean absolute error MAE is 3.79, all of which are the lowest among the four models. Therefore, it can be considered that the CNN model is the most suitable method for predicting the soil organic matter content of arable land in a certain city among these methods.

[0134] This application also provides an application scenario, which applies the above-mentioned soil organic matter prediction mapping method based on deep learning. Specifically: The soil organic matter prediction mapping method based on deep learning provided in this embodiment can be applied in the scenario of agricultural management and planning. The scenario of agricultural management and planning includes data collection and processing links, soil organic matter prediction links, and agricultural decision-making links. First, starting from the data collection and processing link, through means such as remote sensing technology, meteorological data, and on-site sampling, multi-source information of the target area is obtained, and after preprocessing steps, a data set available for input to the deep learning model is obtained. Then, these data are input into the soil organic matter prediction model based on deep learning. After the calculation and analysis of the model, a prediction distribution map of the soil organic matter is obtained. Finally, these prediction results enter the agricultural decision-making link, providing a scientific basis for farmland management, crop planting planning, fertilizer application strategies, etc. The soil organic matter prediction mapping method based on deep learning provided in this embodiment belongs to the soil organic matter prediction link and plays a key role in the process of precision agricultural management and planning.

[0135] The present application has the following technical effects:

[0136] (1) Multi-modal data fusion and feature extraction. Based on multi-modal environmental factors such as remote sensing images, terrain variables, soil types, and land cover data, a unified multi-variable input framework is constructed. The CNN model is used to automatically extract the feature associations between multi-modal data, breaking through the manual dependence of traditional methods on variable selection and combination.

[0137] (2) Utilization of spatial context information. Neighborhood environmental variables of 15×15 pixels are extracted around the sampling points to construct a three-dimensional input tensor (spatial dimension × environmental factor dimension), realizing multi-level feature extraction from local to global and effectively making up for the deficiency of traditional machine learning models in utilizing spatial context information.

[0138] (3) Solved the problem of soil property prediction for highly heterogeneous and fragmented cultivated land. Through multi-layer convolution and pooling operations, the CNN model can capture subtle spatial variation features in fragmented areas. This method is particularly suitable for complex environments such as urban cultivated land, and the accuracy is significantly better than that of traditional machine learning methods.

[0139] In summary, by introducing the convolutional neural network (CNN) model, the present application significantly improves the accuracy and adaptability of soil property mapping. The CNN can automatically extract the spatial context features of multi-modal environmental variables (such as remote sensing images, terrain data, human activities, etc.), and is particularly suitable for dealing with the complex heterogeneity and fragmentation of urban cultivated land. Through multi-layer convolution and pooling operations, the model effectively captures the non-linear relationship between soil properties and environmental factors, realizing high-precision spatial prediction. The coefficient of determination R 2 and the error index are both significantly better than those of traditional machine learning methods. The present application has strong applicability and can be widely applied to soil mapping tasks in complex regions, providing an efficient technical support for precision agriculture and ecological management.

[0140] Based on the same inventive concept, the embodiment of the present application also provides a deep learning-based soil organic matter prediction mapping device for implementing the above-mentioned deep learning-based soil organic matter prediction mapping method. The solution provided by this device for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the following deep learning-based soil organic matter prediction mapping device can refer to the limitations on the deep learning-based soil organic matter prediction mapping method in the above text, and will not be repeated here.

[0141] In an exemplary embodiment, as Figure 4 shown, a deep learning-based soil organic matter prediction mapping device is provided, including:

[0142] The remote sensing image and soil sampling point acquisition module 201 is used to acquire the remote sensing image of the soil within the target area and set a preset number of soil sampling points within the target area.

[0143] The environmental data index acquisition module 202 is used to acquire the environmental data indexes of the soil within the target area; the environmental data indexes include topographic factors, vegetation factors, climate factors, and land use factors; the topographic factors are calculated through a digital elevation model and include slope aspect, topographic wetness index, and curvature; the vegetation factors are obtained through normalization based on remote sensing image data and include the normalized difference vegetation index; the climate factors are obtained based on remote sensing images and meteorological data and include land surface temperature, temperature vegetation drought index, and Palmer drought severity index; the land use factors are further calculated based on the interpreted data of the remote sensing image through the land use type map obtained by classification processing and include paddy fields, irrigated land, and dry land.

[0144] The environmental data index extraction module 203 is used to extract the environmental data indexes within the preset range of each soil sampling point respectively to obtain a plurality of three-dimensional environmental tensor data; each three-dimensional environmental tensor data includes the environmental data index vectors of all pixels within the preset range; the environmental data index vector includes the numerical values of all environmental data indexes, and the length of the environmental data index vector is equal to the number of environmental data indexes.

[0145] The soil organic matter content prediction module 204 is used to input the plurality of three-dimensional environmental tensor data into the soil organic matter content prediction model in sequence to obtain a set of predicted values of the soil organic matter content of the soil sampling points within the target area; the soil organic matter content prediction model is obtained by iteratively training a preset deep learning network model using a sample data set; the sample data set includes the historical three-dimensional environmental tensor data of historical soil sampling points and the corresponding measured values of the soil organic matter content.

[0146] The spatial distribution prediction map generation module 205 is used to obtain the spatial distribution prediction map of the soil organic matter content within the target area according to the set of predicted values of the soil organic matter content of the soil sampling points within the target area and the spatial coordinate values of the soil sampling points on the remote sensing image.

[0147] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 5As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store soil organic matter processing data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a method for predicting and mapping soil organic matter based on deep learning.

[0148] Those skilled in the art can understand that Figure 5 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0149] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0150] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0151] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0152] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0153] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0154] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0155] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A soil organic matter prediction and mapping method based on deep learning, characterized in that: The soil organic matter prediction and mapping method based on deep learning includes: Acquire remote sensing images of soil within the target area and set a preset number of soil sampling points within the target area; Obtaining environmental data indicators of soil within the target area; the environmental data indicators include terrain factors, vegetation factors, climate factors and land use factors; the terrain factors are calculated through digital elevation models, including slope aspect, terrain moisture index and curvature; the vegetation factors are calculated through normalization based on remote sensing image data, including normalized vegetation index; the climate factors are obtained based on remote sensing images and meteorological data, including surface temperature, temperature vegetation drought index and Palmer drought severity index; the land use factors are further calculated based on the land use type map obtained through classification processing based on the interpreted data of remote sensing images, including paddy fields, irrigated land and dry land; Extracting environmental data indicators within a preset range of each soil sampling point respectively to obtain a plurality of three-dimensional environmental tensor data; each three-dimensional environmental tensor data includes an environmental data indicator vector of all pixels within the preset range; the environmental data indicator vector includes the values ​​of all environmental data indicators, and the length of the environmental data indicator vector is equal to the number of environmental data indicators; Inputting a plurality of three-dimensional environmental tensor data into a soil organic matter content prediction model in sequence to obtain a set of predicted values ​​of soil organic matter content at soil sampling points within a target area; the soil organic matter content prediction model is obtained by iteratively training a preset deep learning network model using a sample data set; the sample data set includes historical three-dimensional environmental tensor data of historical soil sampling points and corresponding measured values ​​of soil organic matter content; According to the predicted value set of soil organic matter content of soil sampling points within the target area and the spatial coordinate values ​​of the soil sampling points on the remote sensing image, the spatial distribution prediction map of soil organic matter content within the target area is obtained.

2. The soil organic matter prediction and mapping method based on deep learning according to claim 1 is characterized in that: Multiple three-dimensional environmental tensor data are sequentially input into the soil organic matter content prediction model to obtain a set of predicted values ​​of soil organic matter content at soil sampling points within the target area, which also includes: using the minimum-maximum normalization method to normalize each three-dimensional environmental tensor data.

3. The soil organic matter prediction and mapping method based on deep learning according to claim 1 is characterized in that: The soil organic matter content prediction model comprises: an input layer, three convolutional pooling layers, two fully connected layers and an output layer connected in sequence; The input layer is used to receive three-dimensional environment tensor data; The convolution pooling layer is used to extract local features in the input three-dimensional environment tensor data through a convolution operation, and reduce the dimension of the three-dimensional environment tensor data through a pooling operation; The fully connected layer is used to perform dimensionality reduction processing on the output of the convolutional pooling layer; The output layer is used to calculate the predicted value of the organic matter content of each soil sampling point within the target area using a regression algorithm according to the output of the fully connected layer.

4. The soil organic matter prediction and mapping method based on deep learning according to claim 1 is characterized in that: The method for obtaining the measured value of soil organic matter content specifically includes: At each historical soil sampling site, initial soil samples at a depth of 0–20 cm were collected by the quincunx multi-point mixing method; Performing a pretreatment operation on the initial soil sample to obtain a pretreated soil sample; the pretreatment operation includes air-drying the initial soil sample and grinding and sieving it to remove bricks, tiles, stones, and animal and plant residues mixed in the soil sample; Potassium dichromate oxidation-volumetry method was used to determine the actual value of soil organic matter content in pretreated soil samples.

5. The soil organic matter prediction and mapping method based on deep learning according to claim 1 is characterized in that: The calculation formulas for slope aspect, terrain moisture index and curvature in terrain factors are as follows: Among them, Aspect represents the slope direction of the grid cell; z is the elevation, which represents the height of the grid cell; It represents the elevation change rate of the slope of the grid unit in the x direction, where x represents the east-west direction; It represents the elevation change rate of the slope of the grid cell in the y direction, where y represents the north-south direction; TWI represents the terrain wetness index of the grid cell; α represents the catchment area of ​​the grid cell; tanβ represents the tangent value of the slope of the grid cell; Curvature represents the curvature of the grid cell; It is the second-order derivative of the slope of the grid cell in the x direction, indicating the degree of curvature of the terrain of the grid cell in the x direction; It is the second-order derivative of the slope of the grid cell in the y direction, indicating the degree of curvature of the terrain of the grid cell in the y direction.

6. The soil organic matter prediction and mapping method based on deep learning according to claim 1, characterized in that: The calculation formula of the normalized vegetation index in the vegetation factor is: Among them, NDVI represents the normalized vegetation index of the grid cell; NIR represents the reflectance of the near-infrared band of the grid cell; RED represents the reflectance of the red band of the grid cell.

7. The soil organic matter prediction and mapping method based on deep learning according to claim 1 is characterized in that: The calculation formulas for the surface temperature, temperature vegetation drought index and Palmer drought severity index in the climate factors are: Where LST represents the surface temperature of the grid cell; λ represents the effective wavelength of the thermal infrared band of the grid cell; ρ represents the constant of the grid cell; ε represents the surface emissivity of the grid cell; T B represents the brightness temperature of the grid cell; K1 represents the first band radiation constant of the grid cell, with a value of 1; K2 represents the second band radiation constant of the grid cell, with a value of 2; L λ Represents the radiance value of the grid cell; Among them, PDSI represents the temperature vegetation drought index of the grid unit; d represents the actual water supply and demand deviation of the grid unit; PE represents the potential evapotranspiration of the grid unit; R represents the surface runoff of the grid unit; L represents the soil moisture deficit of the grid unit; α1, β and γ represent the first empirical coefficient, the second empirical coefficient and the third empirical coefficient, respectively, which are used to adjust the contribution of different moisture parameters to the humidity index; k represents the climate adaptation coefficient; Where TVDI represents the Palmer Drought Severity Index of the grid cell; LST min is the minimum land surface temperature, indicating the wet boundary of the grid cell; LST max is the maximum land surface temperature, indicating the dry boundary of the grid cell.

8. A soil organic matter prediction and mapping device based on deep learning, characterized in that: The soil organic matter prediction and mapping device based on deep learning is applied to the soil organic matter prediction and mapping method based on deep learning according to any one of claims 1 to 7, and the soil organic matter prediction and mapping device based on deep learning comprises: A remote sensing image and soil sampling point acquisition module is used to acquire remote sensing images of soil within the target area and set a preset number of soil sampling points within the target area; The environmental data index acquisition module is used to obtain environmental data indicators of soil within the target area; the environmental data indicators include terrain factors, vegetation factors, climate factors and land use factors; the terrain factors are calculated by digital elevation model, including slope aspect, terrain moisture index and curvature; the vegetation factors are calculated by normalization based on remote sensing image data, including normalized vegetation index; the climate factors are obtained based on remote sensing images and meteorological data, including surface temperature, temperature vegetation drought index and Palmer drought severity index; the land use factors are further calculated based on the land use type map obtained by classification processing based on the interpreted data of remote sensing images, including paddy fields, irrigated land and dry land; An environmental data index extraction module is used to extract environmental data indexes within a preset range of each soil sampling point to obtain a plurality of three-dimensional environmental tensor data; each three-dimensional environmental tensor data includes an environmental data index vector of all pixels within the preset range; the environmental data index vector includes the values ​​of all environmental data indicators, and the length of the environmental data index vector is equal to the number of environmental data indicators; A soil organic matter content prediction module is used to sequentially input a plurality of three-dimensional environmental tensor data into a soil organic matter content prediction model to obtain a set of predicted values ​​of soil organic matter content at soil sampling points within a target area; the soil organic matter content prediction model is obtained by iteratively training a preset deep learning network model using a sample data set; the sample data set includes historical three-dimensional environmental tensor data of historical soil sampling points and corresponding measured values ​​of soil organic matter content; The spatial distribution prediction map generation module is used to obtain the spatial distribution prediction map of soil organic matter content within the target area based on the soil organic matter content prediction value set of soil sampling points within the target area and the spatial coordinate values ​​of the soil sampling points on the remote sensing image.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the soil organic matter prediction and mapping method based on deep learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the soil organic matter prediction and mapping method based on deep learning described in any one of claims 1 to 7.

Images