A soil organic matter distribution map generation method, device, equipment and medium

Abstract

This invention provides a method, apparatus, device, and medium for generating soil organic matter distribution maps, belonging to the field of agricultural soil environmental monitoring. The method includes: acquiring spectral data, soil organic matter content data, and satellite multi-source feature data of a target area; inverting the spectral data and soil organic matter content data to obtain an initial soil organic matter distribution map; upsampling the initial soil organic matter distribution map to obtain a proxy ground truth value matching the resolution of the satellite multi-source feature data; performing feature extraction processing on the satellite multi-source feature data to obtain target multi-source feature data; performing data regression modeling based on the proxy ground truth value and the target multi-source feature data to obtain a soil organic matter prediction model; and performing prediction processing on the target multi-source feature data based on the soil organic matter prediction model to obtain the target soil organic matter distribution map. This method can improve the accuracy of soil organic matter distribution data inversion.

Description

Technical Field

[0001] This invention relates to the field of agricultural soil environmental monitoring, specifically to a method, apparatus, equipment, and medium for generating soil organic matter distribution maps. Background Technology

[0002] Current soil organic matter monitoring methods rely on field sampling and laboratory analysis. While this approach offers high accuracy, it suffers from sparse sampling, long cycles, and high costs, making it unsuitable for large-scale continuous monitoring. Although remote sensing inversion enables wide-area, non-destructive monitoring, satellite-scale modeling is generally limited by the sparseness and insufficient spatial representativeness of ground-based samples, resulting in scarce training data, significant label noise interference, and weak model generalization ability. Particularly in tropical and subtropical farmland, cloud and rain weather, complex vegetation, and diverse soil types cause significant fluctuations in optical data quality and difficulties in radar feature extraction. Traditional modeling struggles to characterize the spatial heterogeneity of soil organic matter, and its accuracy falls short of practical application requirements.

[0003] Current research attempts to use hyperspectral imaging combined with machine learning or multi-temporal satellite data for inversion, but this still heavily relies on extensive field sampling, resulting in high costs and limited scalability. While UAV hyperspectral imaging can achieve high-precision local inversion, it lacks effective integration with satellite-scale data, limiting results to small areas. Overall, the existing technological system lacks innovative mechanisms for generating high-density proxy ground truth values ​​at low cost and deeply fusing multi-source satellite features under conditions of scarce sampling, making it difficult to balance high accuracy and operability in soil organic matter remote sensing inversion. Summary of the Invention

[0004] To address the aforementioned problems, this application provides a method for generating soil organic matter distribution maps, which improves the accuracy of soil organic matter distribution data retrieval while reducing costs. The technical solution is as follows: In a first aspect, the present invention provides a method for generating a soil organic matter distribution map, comprising: Acquire spectral data, soil organic matter content data, and satellite multi-source feature data of the target area; The initial soil organic matter distribution map was obtained by inverting the spectral data and soil organic matter content data. The initial soil organic matter distribution map was upsampled to obtain a proxy ground truth value that matches the resolution of the satellite multi-source feature data; Feature extraction processing is performed on satellite multi-source feature data to obtain target multi-source feature data; Data regression modeling is performed based on proxy ground truth and target multi-source feature data to obtain a soil organic matter prediction model; Based on the soil organic matter prediction model, the target multi-source feature data is predicted and processed to obtain the target soil organic matter distribution map.

[0005] Combining the first aspect and the above-mentioned implementation methods, in some possible implementation methods, the spectral data and soil organic matter content data are inverted to obtain an initial soil organic matter distribution map, including: Data correction and noise reduction are performed on the spectral data to obtain the target hyperspectral data; The trained inversion network is used to invert the target hyperspectral data and soil organic matter content data to obtain the initial soil organic matter distribution map.

[0006] Combining the first aspect and the above implementation methods, in some possible implementation methods, the initial soil organic matter distribution map is upsampled to obtain a proxy ground truth value that matches the resolution of satellite multi-source feature data, including: The initial soil organic matter distribution map is resampled based on the spatial resolution of satellite multi-source feature data to obtain a soil organic matter distribution map with target resolution. Pixel aggregation and extraction processing is performed on the pixels of the soil organic matter distribution map at the target resolution to obtain the proxy ground truth value that matches the resolution of the satellite multi-source feature data.

[0007] Combining the first aspect and the above implementation methods, in some possible implementation methods, feature extraction processing is performed on the satellite multi-source feature data to obtain target multi-source feature data, including: The satellite optical reflectance data in the multi-source feature data of satellites is subjected to mask screening to obtain the target optical reflectance data; The satellite radar backscatter data in the satellite multi-source feature data is filtered and transformed, and radar feature derivation processing is performed based on the filtered and transformed satellite radar backscatter data to obtain radar feature data. By combining the target's optical reflectivity data with radar signature data, multi-source signature data of the target is obtained.

[0008] Combining the first aspect and the above implementation methods, in some possible implementation methods, a soil organic matter prediction model is obtained by performing data regression modeling based on proxy ground truth and target multi-source feature data, including: Build a training dataset for the model; A soil organic matter prediction model was generated based on the model training dataset.

[0009] Combining the first aspect and the above implementation methods, in some possible implementation methods, the model training dataset is constructed, including: The model training dataset is obtained by combining proxy ground truth data and soil organic matter content data. Soil organic matter prediction models are generated based on the model training set, including: A soil organic matter prediction model is obtained by performing regression modeling based on the training dataset and target multi-source feature data.

[0010] Combining the first aspect and the above implementation methods, in some possible implementation methods, after obtaining the soil organic matter prediction model, the following are included: The soil organic matter prediction model was tested to obtain the test accuracy value; If the test accuracy value is greater than or equal to the preset accuracy threshold, the target multi-source feature data will be predicted based on the soil organic matter prediction model to obtain the target soil organic matter distribution map.

[0011] Secondly, the present invention also provides a soil organic matter distribution map generation device, comprising: The data acquisition unit is used to acquire spectral data, soil organic matter content data, and satellite multi-source feature data of the target area. The data inversion unit is used to invert spectral data and soil organic matter content data to obtain an initial soil organic matter distribution map; The data sampling unit is used to upsample the initial soil organic matter distribution map to obtain the proxy ground truth value that matches the resolution of the satellite multi-source feature data. The feature extraction unit is used to perform feature extraction processing on satellite multi-source feature data to obtain target multi-source feature data; The data modeling unit is used to perform data regression modeling based on proxy ground truth and target multi-source feature data to obtain a soil organic matter prediction model. The data prediction unit is used to predict and process the target multi-source feature data based on the soil organic matter prediction model to obtain the target soil organic matter distribution map.

[0012] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, Memory, used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the soil organic matter distribution map generation method in any of the above implementations.

[0013] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps in the soil organic matter distribution map generation method described in any of the above implementations.

[0014] The beneficial effects of this invention are as follows: First, by inverting soil organic matter content data with hyperspectral data from UAVs, an initial soil organic matter distribution map is generated. Then, by upsampling the initial soil organic matter distribution, a high-density proxy ground truth value matching the resolution of satellite data is obtained, replacing the direct reliance on a large number of field samples in existing technologies. This solves the problems of sparse samples, high costs, and uneven spatial coverage in existing solutions. Second, by extracting multi-source features from optical and radar satellite data and fusing them with proxy ground truth values ​​to train the soil organic matter prediction model, the influence of single optical data on the quality fluctuations of clouds, rain, and vegetation cover is reduced, improving the model's ability to characterize the spatial heterogeneity of soil organic matter and overcoming the shortcomings of weak generalization ability and insufficient accuracy of existing satellite inversion models. At the same time, it realizes the effective extension of high-precision local information from UAVs to satellite regional-scale mapping, solving the shortcomings of the limited application scope of existing UAV results and their inability to support wide-area monitoring. Finally, under the premise of controllable sampling costs, the accuracy of soil organic matter distribution data inversion is improved. Attached Figure Description To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0015] Figure 1 A schematic flowchart of the method for generating soil organic matter distribution maps provided by the present invention; Figure 2 For the present invention Figure 1 A schematic diagram of a scenario in an embodiment; Figure 3 For the present invention Figure 1 A schematic diagram of an embodiment of S102; Figure 4 For the present invention Figure 1 A schematic diagram of an embodiment of S103; Figure 5 For the present invention Figure 1 A schematic diagram of an embodiment of S104; Figure 6 For the present invention Figure 1 A schematic diagram of an embodiment of S106; Figure 7 A schematic diagram of the soil organic matter distribution map generation device provided by the present invention; Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0017] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0018] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0019] For ease of understanding, the following explains some key terms in this embodiment: Spectral data refers to information about electromagnetic radiation reflected, absorbed, or emitted by an object at different wavelengths or frequencies.

[0020] Soil organic matter content refers to the total amount of various organic substances in the soil, and is a key indicator for measuring soil fertility, carbon storage, and ecological health. This data is usually obtained by measuring collected soil samples through laboratory chemical analysis methods (such as the potassium dichromate oxidation method).

[0021] Satellite multi-source feature data refers to remote sensing data acquired from different satellite platforms with different spatial, spectral, or temporal resolutions, and the various features extracted after processing.

[0022] An initial soil organic matter distribution map is a spatially accurate image of soil organic matter distribution obtained by inverting high-resolution spectral data of a local area with corresponding soil organic matter content data. This distribution map typically covers a small area.

[0023] The surrogate ground truth refers to a high spatial density soil organic matter reference value obtained by upsampling a high-resolution initial soil organic matter distribution map, which matches the resolution of satellite multi-source feature data. This surrogate ground truth, to some extent, compensates for the sparsity of traditional ground truth samples and provides dense labeled data for training satellite-scale models.

[0024] Soil organic matter prediction models are mathematical models obtained through data regression modeling that can predict soil organic matter content based on remote sensing feature data.

[0025] A target soil organic matter distribution map refers to a spatial distribution image of soil organic matter covering the target area with a specific spatial resolution, obtained by predicting and processing multi-source feature data of the target area through a soil organic matter prediction model.

[0026] In traditional soil organic matter remote sensing inversion techniques, ground sample collection relies on field sampling combined with laboratory chemical analysis, resulting in sparse and spatially uneven sampling points. Existing satellite-scale inversion models suffer from limited generalization ability due to insufficient training data and significant label noise. This is particularly true in tropical and subtropical farmland ecosystems, where frequent cloud and rain weather, dynamic changes in vegetation cover, and soil type diversity contribute to fluctuations in optical remote sensing data quality and difficulties in extracting radar backscattering features. Consequently, these models cannot effectively characterize the spatial heterogeneity of soil organic matter, thus affecting the reliability and applicability of the inversion results.

[0027] For example, when monitoring soil organic matter in typical tropical farmland areas of Southeast Asia, optical and radar data are fused and inverted. This region is constantly influenced by the monsoon climate, and cloud cover limits the effective acquisition window for optical data. Furthermore, changes in vegetation indices due to crop growth cycles make it difficult to consistently meet the conditions for bare soil screening. Ground-based measured samples are only distributed at local farmland grid points, resulting in insufficient spatial representativeness. This leads to systematic biases in the inversion model at the regional scale, failing to accurately reflect the actual spatial distribution pattern of soil organic matter, and consequently interfering with the agricultural management decision-making process.

[0028] To address the aforementioned technical problems, this invention provides a method for generating soil organic matter distribution maps. Please refer to... Figure 1 , Figure 1 This is a schematic flowchart of a method for generating a soil organic matter distribution map provided in an embodiment of this application. Figure 1 As shown, the method in this application embodiment may include the following steps S101-S106: S101, acquire spectral data, soil organic matter content data, and satellite multi-source feature data of the target area.

[0029] In this embodiment, the spectral data of the target area is acquired by a hyperspectral sensor mounted on a drone, with a spatial resolution typically reaching the centimeter level. When collecting soil organic matter content data for the target area, it can be obtained by having staff collect soil samples from a small number of ground sampling points in the target area, followed by laboratory chemical analysis of the soil samples. For example, the soil organic matter content can be determined using the potassium dichromate external heating method or an elemental analyzer.

[0030] The satellite multi-source feature data comes from multi-source satellite remote sensing data, which includes optical satellite data and radar satellite data. Among them, optical satellite data is used to obtain the optical band reflectivity characteristics of the target area, and radar satellite data is used to obtain the backscattering intensity characteristics of the target area.

[0031] In one feasible implementation, a drone equipped with a hyperspectral imager is used to cover the target area under clear, cloudless weather conditions, a flight altitude of 80–120 m, a ground resolution of 3–5 cm, and a flight speed of 5–8 m / s, acquiring spectral data of the target area. During the flight, the solar altitude angle, GPS coordinates, and attitude data are recorded simultaneously.

[0032] A checkerboard sampling method was used, with a grid spacing of 20 m × 20 m set in the target area. 30–50 topsoil samples were collected. The precise coordinates of each sampling point were recorded using a GPS locator. The collected soil samples were placed in sealed bags, and after natural air drying, removal of gravel and organic debris, grinding, sieving, and quartering, the soil organic matter content data were obtained in the laboratory.

[0033] Acquire optical and radar satellite data that are closest to the flight date of the UAV, and the optical and radar satellite data together constitute multi-source satellite feature data.

[0034] S102, the spectral data and soil organic matter content data are inverted to obtain the initial soil organic matter distribution map.

[0035] In this embodiment of the application, a trained inversion network is invoked to convert high-resolution spectral data into initial soil organic matter distribution maps at the centimeter or sub-meter level that cover a local area.

[0036] In one feasible implementation, the spectral data of the acquired target area are sequentially subjected to radiometric calibration and atmospheric correction, geometric fine correction, and smoothing filtering preprocessing, and the first-order differential spectrum and vegetation / soil index are calculated.

[0037] Pixel spectral curves corresponding to the coordinates of ground sampling points are extracted, and these curves are paired with measured soil organic matter content data to construct a training dataset. The training dataset is then input into a pre-defined inversion network for training, and the optimal model weights are saved after training. The trained inversion network is then used to perform forward prediction on each pixel within the entire UAV flight area, outputting the predicted soil organic matter value for each pixel and generating an initial soil organic matter distribution map at the centimeter level.

[0038] S103, the initial soil organic matter distribution map is upsampled to obtain the proxy ground truth value that matches the resolution of the satellite multi-source feature data.

[0039] In this embodiment of the application, the upsampling process is intended to convert the initial high-resolution soil organic matter distribution map into the same spatial resolution as the satellite multi-source feature data.

[0040] In one feasible implementation, the initial soil organic matter distribution map is resampled to a pixel resolution that is perfectly aligned with the satellite multi-source feature data by using bilinear interpolation or area-weighted averaging. For each pixel, the arithmetic mean of all centimeter-level pixels or the center pixel value within that pixel is extracted as the proxy ground truth value for that pixel, thereby obtaining the proxy ground truth value covering the entire target area.

[0041] For example, if the initial distribution map has a resolution of 1 meter, while the satellite multi-source feature data has a resolution of 10 meters, the initial distribution map with a resolution of 1 meter can be upsampled to a resolution of 10 meters using methods such as bilinear interpolation, nearest neighbor interpolation, or cubic convolution interpolation. This upsampling process allows high-resolution local soil organic matter information to be effectively expanded and integrated, thereby providing dense, spatially representative label data for satellite-scale modeling.

[0042] S104 performs feature extraction processing on the satellite multi-source feature data to obtain target multi-source feature data.

[0043] In one feasible implementation, for optical satellite data within the multi-source satellite feature data, reflectivity values ​​for different spectral bands can be extracted, or spectral indices such as the normalized vegetation index and bare soil index can be calculated. For radar satellite data within the multi-source satellite feature data, VV and VH polarization backscattering coefficients can be extracted, or derived features such as their ratios and differences can be calculated. By integrating and processing the extracted features, the target multi-source feature data can be obtained.

[0044] S105, based on the proxy ground truth and target multi-source feature data, data regression modeling is performed to obtain the soil organic matter prediction model.

[0045] In this embodiment, the regression modeling process aims to establish a quantitative relationship between surrogate ground truth and target multi-source feature data. For example, multiple linear regression models, support vector regression models, random forest regression models, or gradient boosting tree models can be constructed. During the modeling process, a portion of the surrogate ground truth and the corresponding target multi-source feature data are typically used as a training set. The model parameters are adjusted through optimization algorithms so that the model can accurately predict soil organic matter content.

[0046] S106. Based on the soil organic matter prediction model, the target multi-source feature data is predicted and processed to obtain the target soil organic matter distribution map.

[0047] In one feasible implementation, a trained soil organic matter prediction model is applied to satellite multi-source feature data covering the entire target area. For each pixel within the target area, the soil organic matter prediction model outputs a predicted soil organic matter content value based on its corresponding target multi-source feature data. By predicting all pixels, a high-resolution spatial distribution map of soil organic matter covering the entire target area is finally generated.

[0048] Please refer to the following: Figure 2 , Figure 2 This is the present invention. Figure 1 A schematic diagram of a scenario from an embodiment. (As shown) Figure 2 As shown, an initial soil organic matter distribution map at the centimeter level is generated by inverting UAV hyperspectral data and ground-measured soil organic matter content data. Then, this high-resolution distribution map is upsampled to obtain a proxy ground truth value that matches the resolution of the satellite multi-source feature data. Simultaneously, feature extraction is performed on the satellite multi-source feature data composed of optical and radar data to obtain target multi-source feature data. Subsequently, a soil organic matter prediction model is trained using the proxy ground truth value as the label and the target multi-source feature data as input. Finally, the soil organic matter prediction model is applied to the target multi-source feature data to generate a target soil organic matter distribution map covering the entire target area, achieving an effective extension from local high-resolution information to regional satellite-scale mapping.

[0049] In summary, this application's embodiments invert soil organic matter content data with UAV hyperspectral data to generate an initial soil organic matter distribution map. Then, by upsampling the initial soil organic matter distribution, a high-density proxy ground truth value matching the satellite data resolution is obtained. This replaces the direct reliance on extensive field sampling in existing technologies, solving the problems of sparse samples, high costs, and uneven spatial coverage in existing schemes. Secondly, by extracting multi-source features from optical and radar satellite data and fusing the proxy ground truth value to train the soil organic matter prediction model, the influence of single optical data on quality fluctuations caused by clouds, rain, and vegetation cover is reduced, improving the model's ability to characterize the spatial heterogeneity of soil organic matter and overcoming the shortcomings of weak generalization ability and insufficient accuracy in existing satellite inversion models. Simultaneously, it effectively extends high-precision local information from UAVs to satellite regional-scale mapping, solving the limitations of existing UAV results in application scope and inability to support wide-area monitoring. Ultimately, under the premise of controllable sampling costs, the accuracy of soil organic matter distribution data inversion is improved.

[0050] In some embodiments described above, an inversion process using spectral data and soil organic matter content data is proposed to obtain an initial soil organic matter distribution map. However, in practical applications, the original spectral data may be affected by various environmental factors such as atmospheric effects and sensor noise. Direct inversion processing may lead to insufficient accuracy and reliability of the inversion results, thus affecting the accuracy of the final soil organic matter distribution map. Therefore, please refer to... Figure 3 , Figure 3 This is the present invention. Figure 1 A schematic diagram of an embodiment of S102. (See attached diagram.) Figure 3 As shown, the method in this application embodiment may include the following steps S301-S302: S301 performs data correction and noise reduction on the spectral data to obtain the target hyperspectral data.

[0051] In the embodiments of this application, data correction aims to eliminate systematic errors in spectral data caused by the external environment or the sensor itself, such as atmospheric scattering, absorption effects, and radiometric calibration deviations of the sensor.

[0052] In one feasible implementation, based on a radiative transfer model, the influence of the atmosphere on surface reflectance is eliminated by inputting atmospheric parameters, converting apparent reflectance into true surface reflectance. The corrected spectral data is then subjected to noise reduction processing; for example, a smoothing filter can be applied, setting the polynomial order to 3 and the window size to 11, to smooth the spectral curves of each band, thereby removing high-frequency noise and obtaining the target hyperspectral data.

[0053] S302, the trained inversion network is called to perform inversion processing on the target hyperspectral data and soil organic matter content data to obtain the initial soil organic matter distribution map.

[0054] In this embodiment, the inversion network is a computational model capable of learning the complex nonlinear mapping relationship between spectral data and soil organic matter content. Before being invoked, this network has been trained using a large number of known soil organic matter content samples and their corresponding spectral data, thus enabling it to accurately predict soil organic matter content from spectral features. This inversion network can be implemented using various machine learning or deep learning models; for example, it can be a support vector regression-based model, a random forest regression model, an artificial neural network model, or a more complex convolutional neural network or recurrent neural network model.

[0055] It should be noted that the training process of the inversion network is as follows: extract the pixel spectral curves corresponding to the coordinates of the ground sampling points, pair the extracted pixel spectral curves with the measured soil organic matter content data, construct a training dataset, and divide the training dataset into a training set and a test set according to a preset ratio.

[0056] The pixel spectral curves in the training set are used as input to the inversion network, and the soil organic matter content data in the training set are used as the expected output of the inversion network. The inversion network is trained iteratively. In each iteration, the inversion network calculates the predicted value of soil organic matter content corresponding to the input pixel spectral curve based on the current model weights, and uses the loss function to calculate the error between the predicted value of soil organic matter content and the expected output. Then, based on this error, the model weights of the inversion network are updated through the backpropagation algorithm. When the number of iterations reaches the preset number of iterations or the error of the loss function meets the stopping condition of the early stopping mechanism, the training stops and the optimal model weights are saved, thus completing the training process of the inversion network.

[0057] The number of input layer nodes in the inversion network is the number of preprocessed bands. The hidden layer adopts a 3-5 layer fully connected structure with a neuron arrangement of 128-64-32-16 and an activation function of ReLU. The output layer is a single neuron used to predict soil organic matter content. The training parameters are as follows: Adam optimizer is used with an initial learning rate of 0.001, a batch size of 16, a loss function of mean squared error, 200-500 training epochs, and an early stopping mechanism with a patience value of 30.

[0058] In one feasible implementation, the inversion network receives target hyperspectral data as input, performs feature extraction and nonlinear mapping through its internal convolutional layers, pooling layers and fully connected layers, and finally outputs the predicted value of soil organic matter content for each pixel, thereby generating an initial soil organic matter distribution map.

[0059] In summary, this application effectively eliminates or reduces interference from external factors such as the atmosphere and sensors on spectral signals by performing data correction and noise reduction on the spectral data, ensuring higher purity and accuracy of the spectral data input to the inversion network. Secondly, by calling the trained inversion network for inversion processing, the model can fully learn and capture the complex nonlinear relationship between spectral features and soil organic matter content, avoiding the limitations that may exist in traditional linear or simple models. This combination of high-quality input data and intelligent inversion model significantly improves the accuracy and reliability of the initial soil organic matter distribution map, providing a more solid and accurate foundation for subsequent soil organic matter prediction models, thereby improving the overall quality of the final soil organic matter distribution map.

[0060] In practice, initial soil organic matter distribution maps typically have low spatial resolution. Directly performing simple upsampling may lead to data distortion or an inability to accurately match the fine scale of satellite multi-source feature data, thus affecting the accuracy of subsequent data regression modeling. Therefore, please refer to [link to relevant documentation]. Figure 4 , Figure 4 This is the present invention. Figure 1 A schematic diagram of an embodiment of S103. (See attached diagram.) Figure 4 As shown, the method in this application embodiment may include the following steps S401-S402: S401, based on the spatial resolution of satellite multi-source feature data, the initial soil organic matter distribution map is resampled to obtain the target resolution soil organic matter distribution map.

[0061] In this embodiment, resampling is a technique for altering the spatial resolution of an image or raster data. Its function is to transform the initial soil organic matter distribution map from its original resolution to an intermediate resolution, i.e., a target resolution soil organic matter distribution map. This intermediate resolution is correlated with the spatial resolution of satellite multi-source feature data, laying the foundation for subsequent pixel aggregation and extraction processing. Resampling can be implemented using various interpolation algorithms. For example, nearest neighbor interpolation can be used, directly assigning the value of the nearest neighbor pixel to the new pixel; bilinear interpolation can be used, determining the new pixel value through the weighted average of the surrounding four pixels; or cubic convolution interpolation can be used, calculating using the weighted average of the surrounding 16 pixels to obtain a smoother interpolation result.

[0062] For example, the initial soil organic matter distribution map obtained through inversion has a spatial resolution of 100 meters, while the spatial resolution of the acquired satellite multi-source feature data is 10 meters. To generate a surrogate ground truth that matches the resolution of the satellite multi-source feature data, the initial soil organic matter distribution map can be resampled based on the 10-meter spatial resolution of the satellite multi-source feature data. For instance, a cubic convolution interpolation algorithm can be used to resample the initial soil organic matter distribution map at a 100-meter resolution to a target resolution soil organic matter distribution map at 10 meters.

[0063] S402, pixel aggregation and extraction processing is performed on the pixels of the soil organic matter distribution map at the target resolution to obtain the proxy ground truth value that matches the resolution of the satellite multi-source feature data.

[0064] In this embodiment, pixel aggregation extraction processing refers to merging multiple pixels in the target resolution soil organic matter distribution map into a new pixel and calculating the value of the new pixel, thereby further adjusting the spatial scale of the data. Its purpose is to ensure that the final generated surrogate ground truth value completely matches the satellite multi-source feature data in terms of spatial resolution, eliminating inconsistencies caused by resolution differences. The surrogate ground truth value refers to the soil organic matter distribution data that, after a series of processing steps, completely matches the satellite multi-source feature data in terms of spatial resolution.

[0065] For example, using a 10-meter pixel from satellite multi-source feature data as a benchmark, the average value of all 10-meter pixels within the corresponding area in the target resolution soil organic matter distribution map is calculated to obtain a new 10-meter pixel value representing that area. In this way, the final surrogate ground truth value will perfectly match the satellite multi-source feature data in spatial resolution, with each surrogate ground truth pixel precisely corresponding to a satellite multi-source feature data pixel.

[0066] In summary, this application, through step-by-step resampling and pixel aggregation extraction, not only elevates the low-resolution initial soil organic matter distribution map to a finer scale matching high-resolution satellite multi-source feature data, but also ensures that the generated surrogate ground truth values ​​are highly consistent with the satellite multi-source feature data in both spatial location and numerical value. This precise spatial matching provides high-quality, highly reliable training samples for subsequent data regression modeling based on the surrogate ground truth values ​​and target multi-source feature data, thereby significantly improving the training effect of the soil organic matter prediction model and the accuracy and fineness of the final generated target soil organic matter distribution map. This makes the prediction results closer to reality and has higher application value.

[0067] In some embodiments described above in this application, a method for generating soil organic matter distribution maps is proposed, which involves feature extraction processing of satellite multi-source feature data to obtain target multi-source feature data. However, satellite multi-source feature data typically contains various types of data, such as optical data and radar data, which have different physical characteristics and noise patterns. Without targeted processing and effective fusion, the extracted feature information may be incomplete or redundant, thus affecting the accuracy and robustness of subsequent soil organic matter prediction models. Based on this, please refer to... Figure 5 , Figure 5 This is the present invention. Figure 1 A schematic diagram of an embodiment of S104. (See attached diagram.) Figure 5 As shown, the method in this application embodiment may include the following steps S501-S503: S501 performs masking and filtering on the satellite optical reflectivity data in the satellite multi-source feature data to obtain the target optical reflectivity data.

[0068] In this embodiment of the application, the satellite optical reflectance data is subjected to masking and screening processing to remove non-target areas or outliers in the satellite optical reflectance data, such as clouds, water bodies, buildings or high-density vegetation, so as to ensure that the optical data used in subsequent analysis can accurately reflect the spectral characteristics of the soil itself.

[0069] S502 filters and transforms the satellite radar backscatter data in the satellite multi-source feature data, and performs radar feature derivation processing based on the filtered and transformed satellite radar backscatter data to obtain radar feature data.

[0070] In this embodiment, satellite radar backscatter data undergoes filtering and transformation processing to reduce speckle noise, a common feature in radar data, and convert it into parameters with greater physical meaning. Filtering can employ spatial domain filtering algorithms, such as Lee filtering and Gamma-MAP filtering, to suppress noise and preserve image details. Transformation can convert the original backscatter intensity values ​​to decibel (dB) values ​​or perform polarization decomposition to obtain different polarization parameters, thereby enhancing the data's sensitivity to ground feature characteristics.

[0071] Radar feature derivation processing based on filtered and transformed satellite radar backscatter data aims to extract features related to soil organic matter from the processed radar data. Derived features may include backscattering coefficients and polarization ratios of different polarization channels, polarization decomposition parameters, and texture features obtained through texture analysis (such as the Gray-Level Co-occurrence Matrix, GLCM). The polarization ratio is expressed as VV / VH, where VV represents co-polarization and VH represents the ratio of co-polarization to cross-polarization.

[0072] S503 combines target optical reflectivity data with radar signature data to obtain target multi-source signature data.

[0073] In this embodiment, target optical reflectance data is combined with radar feature data to fuse data from different sources and of different types, all of which have undergone preprocessing and feature extraction, to form a more comprehensive and information-rich feature set. This combination can fully utilize the sensitivity of optical data to the surface spectral response and the penetrating power of radar data to the surface structure and water content, achieving complementary advantages and thus providing higher-quality input for subsequent soil organic matter prediction models.

[0074] In one feasible implementation, the satellite optical reflectance data, such as optical images from the Sentinel-2 satellite, is subjected to masking and filtering. For example, image processing techniques are used to identify and remove clouds, cloud shadows, water bodies, and high-density vegetation areas from the images. For instance, the Normalized Difference Vegetation Index (NDVI) is calculated and a threshold is set to distinguish between vegetation and bare soil, or existing land cover classification maps are used for masking, thereby obtaining target optical reflectance data that only includes bare soil areas.

[0075] For satellite radar backscatter data, such as C-band VV and VH polarization data from the Sentinel-1 satellite, filtering can be performed first. For example, the Lee filtering algorithm can be used to effectively suppress speckle noise, which is common in radar images, to improve data quality. Subsequently, the filtered radar backscatter intensity values ​​are converted into decibel values ​​for subsequent analysis.

[0076] Building upon this foundation, further radar feature derivation processing can be performed, such as calculating the polarization ratio of different polarization channels or extracting physical parameters like surface scattering and volume scattering using polarization decomposition algorithms. The target optical reflectivity data obtained through this processing is then combined with the radar feature data. For example, this can be achieved through feature stacking, where the two types of data are concatenated at the pixel level to form a comprehensive feature vector containing both spectral and radar structural information, serving as the final multi-source target feature data.

[0077] In summary, this application effectively eliminated interference from non-soil areas by masking satellite optical reflectance data, ensuring the purity of optical features. Simultaneously, filtering, transformation, and feature derivation processing of satellite radar backscatter data overcame the inherent noise problem of radar data and extracted features sensitive to soil physical properties. Finally, combining these two optimized heterogeneous data sources fully leveraged the advantages of optical data in spectral identification and radar data in penetration and structural information, achieving information complementarity and fusion. This resulted in target multi-source feature data with higher information content and stronger representativeness, significantly improving the accuracy, stability, and adaptability to complex environments of subsequent soil organic matter prediction models.

[0078] In one feasible implementation, the step of performing data regression modeling based on surrogate ground truth and target multi-source feature data to obtain a soil organic matter prediction model is further specifically performed as follows: Build a training dataset for the model.

[0079] In the embodiments of this application, constructing a model training dataset refers to preparing a data set for training a machine learning model. The model training dataset typically contains pairs of input features (derived from target multi-source feature data) and corresponding output labels (derived from surrogate ground truth values), and its purpose is to provide the model with samples for learning the mapping relationship between inputs and outputs.

[0080] In one feasible implementation, existing proxy ground truth and target multi-source feature data can be systematically sampled and partitioned into different subsets such as training set, validation set, and test set. Furthermore, data augmentation techniques can be used to expand the diversity and quantity of training data, such as making minor perturbations or transformations to existing data points, to improve the model's generalization ability.

[0081] And a soil organic matter prediction model is generated based on the model training set.

[0082] In this embodiment, generating a soil organic matter prediction model based on a model training set refers to using a pre-constructed model training dataset and a specific machine learning algorithm to learn and build a mathematical model capable of predicting soil organic matter content. Its function is to extract the patterns and regularities contained in the training data, forming a functional entity that can be used for future predictions.

[0083] In one feasible implementation, advanced supervised learning algorithms, such as Random Forest, Gradient Boosting Machines (GBM), or Support Vector Regression (SVR), can be employed to capture complex nonlinear relationships in environmental data. Alternatively, deep learning architectures, such as Convolutional Neural Networks (CNNs) or Recurrent Neural Networks (RNNs), can be utilized, particularly when processing high-dimensional or sequential feature data, as they can automatically extract hierarchical features and learn complex patterns.

[0084] For example, when constructing the model training dataset, the upsampled surrogate ground truth values ​​can be paired with the target multi-source feature data obtained through feature extraction. For instance, for each spatial location, the corresponding surrogate ground truth value is used as the label, and the target multi-source feature data (such as optical reflectivity, radar features, etc.) is used as the input feature. Then, these paired datasets are randomly divided into training, validation, and test sets. When generating a soil organic matter prediction model based on the model training set, a random forest regressor can be used as the prediction model. The target multi-source feature data in the training set is input into the random forest regressor, and the surrogate ground truth value is used as the supervision signal for training. During training, the model constructs multiple decision trees and improves the accuracy and stability of predictions through ensemble learning. After training, the random forest regressor becomes a soil organic matter prediction model that can be used for prediction.

[0085] In summary, the embodiments of this application, through data regression modeling, ensure that the soil organic matter prediction model can be built based on sufficient and high-quality training data. This effectively avoids the problem of poor prediction model performance caused by insufficient data preparation or non-standard model generation process. Therefore, the obtained soil organic matter prediction model has higher accuracy and robustness, resulting in a more accurate and reliable soil organic matter distribution map.

[0086] Please see Figure 6 , Figure 6 This is the present invention. Figure 1 A schematic diagram of an embodiment of S106. (See attached diagram.) Figure 6 As shown, the method in this application embodiment may include the following steps S601-S604: S601, based on the combination of proxy ground truth and soil organic matter content data, to obtain the model training dataset.

[0087] In one feasible implementation, spatial registration is performed between the surrogate ground truth and soil organic matter content data. For example, for each pixel in the surrogate ground truth, if its spatial location coincides with the location of a measured soil organic matter content point or is within a preset neighborhood, then the soil organic matter content value of that measured point is used as the ground truth label for that pixel. For pixels without a corresponding measured point, the value of the surrogate ground truth itself can be used as its ground truth label, or a spatial interpolation method (such as Kriging interpolation, inverse distance weighted interpolation, etc.) can be used to combine neighboring measured points and the surrogate ground truth to determine its label. This results in a sample set containing the features (target multi-source feature data) of each pixel and the corresponding soil organic matter ground truth label, i.e., the combined training dataset.

[0088] S602, based on the training dataset and target multi-source feature data, performs regression modeling to obtain a soil organic matter prediction model.

[0089] In one feasible implementation, a Gradient Boosting Decision Tree (GBDT) is selected as the regression model. The constructed training dataset is input into the regression model for training. The regression model iteratively trains multiple weak learners and combines their predictions to gradually reduce the prediction error. During training, the regression model learns multi-source feature data of the target, such as optical reflectivity, vegetation index, radar backscattering coefficient, etc., and the complex nonlinear relationship between these features and soil organic matter content. After sufficient training, a soil organic matter prediction model is obtained that can predict soil organic matter content based on new multi-source feature data of the target.

[0090] S603, the soil organic matter prediction model was tested to obtain the test accuracy value.

[0091] S604. If the test accuracy value is greater than or equal to the preset accuracy threshold, the target multi-source feature data is predicted based on the soil organic matter prediction model to obtain the target soil organic matter distribution map.

[0092] It should be noted that in practical applications, directly using untested prediction models may result in insufficient accuracy of the generated soil organic matter distribution maps, thus affecting subsequent decision-making and application effectiveness. Therefore, model testing of soil organic matter prediction models is necessary.

[0093] In this embodiment, testing the soil organic matter prediction model to obtain a test accuracy value refers to the process of evaluating the performance of a trained prediction model to quantify its predictive ability and reliability. The test accuracy value is a key indicator for measuring model performance, reflecting the model's generalization ability on unseen data. This test can be performed by dividing a portion of the original dataset into independent test sets, allowing the model to make predictions on these test sets, and then comparing the prediction results with the true values ​​of the test sets. Statistical indicators such as mean squared error (MSE), coefficient of determination (R²), or mean absolute error (MAE) are calculated as the test accuracy value.

[0094] The preset accuracy threshold is a pre-defined standard value used to determine whether the test accuracy of the soil organic matter prediction model has reached an acceptable level. This preset accuracy threshold can be determined based on actual application needs, industry standards, or expert experience. For example, in some agricultural applications, an R² value of 0.7 or higher may be required for the model to be considered usable. Alternatively, it can be set through historical data analysis and comparison with the performance of existing mature models to ensure that the performance of the new model is at least as good as or better than the existing level. The steps for predicting and processing the target multi-source feature data based on the soil organic matter prediction model to obtain the target soil organic matter distribution map are described in the embodiment of step S106 above, and will not be repeated here.

[0095] For example, after obtaining the soil organic matter prediction model, it can be tested using an independent validation dataset. This validation dataset contains data points with known soil organic matter content and their corresponding target multi-source feature data. The soil organic matter prediction model is applied to this validation dataset, and the coefficient of determination (R²) between the predicted and actual values ​​is calculated. For example, an R² value greater than 0.75 can be set as a preset accuracy threshold. If the calculated R² value is 0.82, which is greater than 0.75, the model is considered qualified and can continue to be used to predict the target multi-source feature data of the entire target area to generate the final soil organic matter distribution map. Conversely, if the R² value is only 0.65, it indicates that the model accuracy is insufficient. In this case, further prediction processing can be omitted, and the model should be retrained or its parameters adjusted until the model accuracy meets the requirements.

[0096] In summary, this application combines spatially continuous surrogate ground truth with high-precision discrete soil organic matter content data to construct a training dataset that not only has stronger spatial representativeness but also improves local accuracy, providing a more solid foundation for model training. Based on this, regression modeling is performed using this optimized training dataset and target multi-source feature data, enabling the generated soil organic matter prediction model to more fully learn and utilize multi-source remote sensing information, thereby significantly improving the model's prediction accuracy and generalization ability, ultimately generating a more accurate and reliable soil organic matter distribution map. After generating the soil organic matter prediction model, a model testing and accuracy evaluation process is introduced. This mechanism effectively selects prediction models that meet the accuracy requirements, avoiding the risks associated with directly using low-precision models for prediction. Therefore, the final generated target soil organic matter distribution map has higher accuracy and reliability.

[0097] The following will combine Figure 7 This application provides a detailed description of the soil organic matter distribution map generation device provided in its embodiments. It should be noted that... Figure 7 The soil organic matter distribution map generation device in the present application is used to perform the present application. Figures 1-6 The methods shown in the embodiments are for illustrative purposes only, illustrating the parts relevant to the embodiments of this application. For specific technical details not disclosed, please refer to this application. Figures 2-8 In the embodiment shown, the soil organic matter distribution map generation device 700 may include a data acquisition unit 701, a data inversion unit 702, a data sampling unit 703, a feature extraction unit 704, a data modeling unit 705, and a data prediction unit 706, as detailed below: The data acquisition unit 701 is used to acquire spectral data, soil organic matter content data, and satellite multi-source feature data of the target area. Data inversion unit 702 is used to invert spectral data and soil organic matter content data to obtain an initial soil organic matter distribution map; Data sampling unit 703 is used to upsample the initial soil organic matter distribution map to obtain a proxy ground truth value that matches the resolution of satellite multi-source feature data; The feature extraction unit 704 is used to perform feature extraction processing on satellite multi-source feature data to obtain target multi-source feature data; Data modeling unit 705 is used to perform data regression modeling based on proxy ground truth and target multi-source feature data to obtain a soil organic matter prediction model; The data prediction unit 706 is used to predict and process the target multi-source feature data based on the soil organic matter prediction model to obtain the target soil organic matter distribution map.

[0098] The soil organic matter distribution map generation device 700 provided in the above embodiments can realize the technical solutions described in the above soil organic matter distribution map generation method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above soil organic matter distribution map generation method embodiments, and will not be repeated here.

[0099] like Figure 8 As shown, the present invention also provides an electronic device 800. The electronic device 800 includes a processor 801, a memory 802, and a display 803. Figure 8 Only some components of the electronic device 800 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0100] In some embodiments, processor 801 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 802 or process data, such as the soil organic matter distribution map generation method of the present invention.

[0101] In some embodiments, processor 801 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 801 may be local or remote. In some embodiments, processor 801 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, intranet, multi-cloud, etc., or any combination thereof.

[0102] In some embodiments, memory 802 may be an internal storage unit of electronic device 800, such as a hard disk or memory of electronic device 800. In other embodiments, memory 802 may also be an external storage device of electronic device 800, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 800.

[0103] Furthermore, the memory 802 may include both internal storage units of the electronic device 800 and external storage devices. The memory 802 is used to store application software and various types of data installed on the electronic device 800.

[0104] In some embodiments, display 803 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 803 is used to display information from electronic device 800 and to display a visual user interface. Components 801-803 of electronic device 800 communicate with each other via a system bus.

[0105] In one embodiment, when the processor 801 executes the data monitoring program in the memory 802, the following steps can be implemented: Acquire spectral data, soil organic matter content data, and satellite multi-source feature data of the target area; The initial soil organic matter distribution map was obtained by inverting the spectral data and soil organic matter content data. The initial soil organic matter distribution map was upsampled to obtain a proxy ground truth value that matches the resolution of the satellite multi-source feature data; Feature extraction processing is performed on satellite multi-source feature data to obtain target multi-source feature data; Data regression modeling is performed based on proxy ground truth and target multi-source feature data to obtain a soil organic matter prediction model; Based on the soil organic matter prediction model, the target multi-source feature data is predicted and processed to obtain the target soil organic matter distribution map.

[0106] It should be understood that when the processor 801 executes the data monitoring program in the memory 802, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0107] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 800 mentioned. Electronic device 800 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 800 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0108] Accordingly, this application also provides a computer-readable storage medium for storing a computer-readable program or instruction. When the program or instruction is executed by a processor, it can implement the steps or functions in the soil organic matter distribution map generation method provided in the above-described method embodiments.

[0109] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0110] The method, apparatus, equipment, and medium for generating soil organic matter distribution maps provided by this invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are merely for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will find variations in specific implementation methods and application scope based on the ideas of this invention. In summary, the contents of this specification should not be construed as limiting the invention.

Claims

1. A method for generating a soil organic matter distribution map, characterized in that, include: Acquire spectral data, soil organic matter content data, and satellite multi-source feature data of the target area; The initial soil organic matter distribution map is obtained by inverting the spectral data and the soil organic matter content data. The initial soil organic matter distribution map is upsampled to obtain a proxy ground truth value that matches the resolution of the satellite multi-source feature data; Feature extraction processing is performed on the satellite multi-source feature data to obtain target multi-source feature data; Based on the proxy ground truth and the target multi-source feature data, a data regression model is performed to obtain a soil organic matter prediction model. Based on the soil organic matter prediction model, the target multi-source feature data is predicted and processed to obtain the target soil organic matter distribution map.

2. The method according to claim 1, characterized in that, The process of inverting the spectral data and the soil organic matter content data to obtain an initial soil organic matter distribution map includes: The spectral data is subjected to data correction and data noise reduction processing to obtain the target hyperspectral data; The trained inversion network is invoked to perform inversion processing on the target hyperspectral data and the soil organic matter content data to obtain an initial soil organic matter distribution map.

3. The method according to claim 1, characterized in that, The upsampling process of the initial soil organic matter distribution map to obtain a proxy ground truth value that matches the resolution of the satellite multi-source feature data includes: Based on the spatial resolution of the satellite multi-source feature data, the initial soil organic matter distribution map is resampled to obtain a soil organic matter distribution map with target resolution. Pixel aggregation and extraction processing is performed on the pixels of the soil organic matter distribution map at the target resolution to obtain a proxy ground truth value that matches the resolution of the satellite multi-source feature data.

4. The method according to claim 1, characterized in that, The step of performing feature extraction processing on the satellite multi-source feature data to obtain target multi-source feature data includes: The satellite optical reflectance data in the satellite multi-source feature data is subjected to mask screening processing to obtain the target optical reflectance data; The satellite radar backscatter data in the satellite multi-source feature data is filtered and transformed, and radar feature derivation processing is performed based on the filtered and transformed satellite radar backscatter data to obtain radar feature data. The target optical reflectivity data is combined with the radar feature data to obtain target multi-source feature data.

5. The method according to claim 1, characterized in that, The process of performing data regression modeling based on the proxy ground truth and the target multi-source feature data to obtain a soil organic matter prediction model includes: Build a training dataset for the model; A soil organic matter prediction model is generated based on the training dataset of the model.

6. The method according to claim 5, characterized in that, The dataset used to build the model training dataset includes: The model training dataset is obtained by combining the proxy ground truth values ​​and the soil organic matter content data. The process of generating a soil organic matter prediction model based on the model training dataset includes: A soil organic matter prediction model is obtained by performing regression modeling based on the training dataset and the target multi-source feature data.

7. The method according to claim 5, characterized in that, After generating the soil organic matter prediction model based on the model training dataset, the method further includes: The soil organic matter prediction model was tested to obtain the test accuracy value; If the test accuracy value is greater than or equal to the preset accuracy threshold, then the target multi-source feature data is predicted based on the soil organic matter prediction model to obtain the target soil organic matter distribution map.

8. A soil organic matter distribution map generation device, characterized in that, The device includes: The data acquisition unit is used to acquire spectral data, soil organic matter content data, and satellite multi-source feature data of the target area. The data inversion unit is used to invert the spectral data and the soil organic matter content data to obtain an initial soil organic matter distribution map; The data sampling unit is used to perform upsampling processing on the initial soil organic matter distribution map to obtain a proxy ground truth value that matches the resolution of the satellite multi-source feature data; The feature extraction unit is used to perform feature extraction processing on the satellite multi-source feature data to obtain target multi-source feature data; The data modeling unit is used to perform data regression modeling based on the proxy ground truth and the target multi-source feature data to obtain a soil organic matter prediction model. The data prediction unit is used to perform prediction processing on the target multi-source feature data based on the soil organic matter prediction model to obtain the target soil organic matter distribution map.

9. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable program code; A processor for calling and running the executable program code from the memory, causing the electronic device to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the method as described in any one of claims 1 to 7.

Images