Crop residue coverage estimation method based on fusion of optical spectral index and sar remote sensing image

Abstract

This invention discloses a method for estimating crop straw coverage based on the fusion of optical spectral index and SAR remote sensing imagery, comprising the following steps: Step 1, data acquisition; Step 2, data preprocessing; Step 3, feature variable extraction; Step 4, variable screening and importance ranking; Step 5, collaborative inversion model construction; Step 6, regional straw coverage mapping. This invention overcomes the shortcomings of single optical remote sensing (susceptibility to cloud and rain interference and data gaps) and single SAR remote sensing (difficult interpretation and low accuracy of individual inversion) by synergistically utilizing the spectral absorption characteristics of optical remote sensing imagery and the backscattering characteristics of SAR radar imagery. It obtains the optimal variable combination through a variable selection strategy based on random forest importance ranking, ensuring the model's efficiency and robustness. The inversion model is constructed using a nonlinear machine learning algorithm, solving the problem of insufficient fitting ability of linear models, enabling continuous and stable monitoring over all weather conditions, with high precision and a large area.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing monitoring technology, specifically a method for estimating crop straw coverage based on the fusion of optical spectral index and SAR remote sensing imagery. Background Technology

[0002] Crop straw coverage is a core quantitative indicator of the effectiveness of conservation tillage. Accurate estimation of this indicator can directly reflect the straw coverage status of farmland, providing key data for black soil protection, farmland degradation prevention and control, and soil wind and water erosion control. At the same time, it can support government departments in accurately grasping the area of ​​conservation tillage promotion, determining the amount of agricultural subsidies, and scientifically guiding agricultural production and ecological farmland construction, which is of great practical significance for ensuring farmland quality and improving the regional agricultural ecological environment.

[0003] Current methods for obtaining crop straw coverage mainly fall into four categories: traditional field measurement, single optical remote sensing inversion, single SAR remote sensing inversion, and laboratory fluorescence methods. Each has the following drawbacks: 1. Traditional field measurement relies on transect or photographic methods, resulting in low efficiency. It requires significant manpower, material resources, and financial investment, is time-consuming and labor-intensive, and cannot complete large-scale surveys quickly. Furthermore, it only acquires discrete sampling point data, lacking the ability to monitor large-scale continuous spatial distribution, making it difficult to reflect the overall regional situation. Moreover, the measurement results are affected by human operations and subjective judgments such as the tension of the rope and the interpretation of photographs, leading to poor data objectivity and consistency. 2. Single optical remote sensing inversion relies on Landsat... 8. Sentinel-2 and other satellite imagery, based on the spectral absorption characteristics of straw cellulose and lignin to construct index inversions such as NDTI and STI, is currently the mainstream technology. However, it is highly susceptible to weather conditions such as clouds, rain, and fog. Optical signals cannot penetrate cloud layers, and data loss often occurs during the critical period of straw monitoring after autumn harvest due to cloudy and rainy conditions. At the same time, its reflectivity is easily affected by soil type, roughness, moisture content, and background interference from residual green vegetation in the field. In areas with large variations in soil moisture and similar spectral characteristics of straw and soil, the inversion accuracy will drop significantly, and the model's universality and stability are insufficient; 3. Single SAR remote sensing inversion utilizes synthetic aperture radar data such as Sentinel-1. While it offers advantages for all-day, all-weather monitoring, its radar backscattering coefficient is heavily influenced by multiple factors, including surface roughness, dielectric constant, straw coverage, soil and straw moisture content, radar incident angle, and polarization. This makes interpretation extremely difficult, and relying solely on single radar data cannot accurately extract straw coverage signals. The inversion accuracy is significantly lower than that of optical remote sensing, with large estimation biases, making it difficult to achieve high-precision monitoring independently. 4. The fluorescence method relies on the difference in fluorescence between straw and soil excited by ultraviolet light to distinguish them. However, this method remains only at the laboratory research stage. In the natural light environment of the field, the fluorescence signal is weak and difficult to detect, and there is a lack of high-energy excitation energy that can be applied over a large area, making it unsuitable for large-scale practical monitoring in the field. Summary of the Invention

[0004] The purpose of this invention is to provide a method for estimating crop straw coverage based on the fusion of optical spectral index and SAR remote sensing imagery, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for estimating crop straw coverage based on fusion of optical spectral index and SAR remote sensing imagery, comprising the following steps: Step 1, data acquisition; Step 2, data preprocessing; Step 3, feature variable extraction; Step 4, variable screening and importance ranking; Step 5, collaborative inversion model construction; Step 6, regional straw coverage mapping.

[0006] In step one above, optical remote sensing images and SAR remote sensing images of the target monitoring area after crop harvest are acquired simultaneously, straw coverage data of sampling points in the target monitoring area are collected, and the center latitude and longitude coordinates of the sampling points are recorded to construct a ground-based measured dataset.

[0007] In step two above, the SAR remote sensing images acquired in step one are subjected to orbit correction, thermal noise removal, radiometric calibration, multi-view processing, filtering, terrain correction, decibel reduction, and resampling; atmospheric correction is performed on the optical remote sensing images, converting the top atmospheric reflectance to the surface reflectance, and band resampling is performed to ensure that the spatial resolution and projection coordinate system of the optical remote sensing images are consistent with those of the SAR remote sensing images.

[0008] In step three above, spectral indices sensitive to straw are extracted based on the preprocessed optical remote sensing images, and VV polarization backscattering coefficients and VH polarization backscattering coefficients of the corresponding sampling points are extracted based on the preprocessed SAR remote sensing images, thus completing the extraction of multi-source feature variables.

[0009] In step four above, correlation analysis is used to remove feature variables that are weakly correlated with straw coverage, and then the variables are sorted according to their importance to select the optimal combination of feature variables.

[0010] In step five above, the optimal combination of feature variables selected in step four is used as the input variable, and the measured straw coverage on the ground is used as the output variable. The nonlinear collaborative estimation model of crop straw coverage is constructed and trained using machine learning algorithms. The prediction error is minimized by optimizing the model parameters. After the model training is completed, the optimal model is selected by evaluating the model accuracy.

[0011] In step six above, the optimal model selected in step five is used to perform pixel-by-pixel calculations on the remote sensing images of the entire monitoring area to generate a spatial distribution map of crop straw coverage at the regional scale, and coverage classification is completed according to a preset threshold.

[0012] In step one, the preferred optical remote sensing image is Sentinel-2 MSI image, which covers the visible light, near-infrared and short-wave infrared bands; the preferred SAR remote sensing image is Sentinel-1 satellite image data, which is in interferometric wide-swath mode, GRD product level, and VV and VH dual polarization, where VV is vertical transmission and vertical reception, and VH is vertical transmission and horizontal reception.

[0013] In step one, the straw coverage data is calculated using either the transect method or the photographic method.

[0014] In step two, orbit correction involves correcting satellite orbit information using orbital ephemeris files; thermal noise removal eliminates thermal noise generated by the radar system; radiometric calibration converts radar image pixel values ​​into radar backscattering coefficients; multi-view processing performs multi-view processing on the image to suppress speckle noise; filtering removes speckle noise while preserving edge information, preferably using a Refined Lee filter for adaptive filtering; terrain correction uses a digital elevation model for range-Doppler terrain correction to eliminate geometric distortion caused by terrain undulations; decibel conversion converts the backscattering coefficients in linear units to logarithmic units; and resampling resamples the SAR remote sensing image to a uniform spatial resolution.

[0015] In step three, the spectral indices include NDI5, NDI7, NDSVI, SRNDI, NDRI, NDTI, and STI, preferably including NDTI and STI. The calculation formula is as follows:

[0016]

[0017]

[0018] in, The normalized differential tillage index. For simple farming index, It is the first band of shortwave infrared, with a center wavelength of 1610 nm. It is the second band of shortwave infrared, with a center wavelength of 2190nm.

[0019] In step four, the correlation analysis uses the Pearson correlation coefficient, and the ranking is achieved using the variable importance assessment function of the random forest algorithm.

[0020] In step four, the optimal combination of feature variables selected is STI, NDTI, VH polarization backscattering coefficient and VV polarization backscattering coefficient.

[0021] In step five, the machine learning algorithm used is either a backpropagation neural network or a random forest.

[0022] The BP neural network model is constructed as follows: the number of input layer nodes corresponds to the number of feature variables, the hidden layer is processed by a transfer function, the output layer outputs the predicted coverage value, and the weights are adjusted by the backpropagation algorithm to minimize the prediction error.

[0023] The random forest construction model is as follows: construct multiple decision trees, use Bootstrap sampling to generate a training set, grow decision trees by splitting nodes, and finally obtain the predicted value by averaging the regression results of multiple trees.

[0024] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention fully leverages the complementary advantages of the two types of remote sensing data by synergistically utilizing the spectral absorption characteristics of optical remote sensing images and the backscattering characteristics of SAR radar images, thereby achieving continuous and stable monitoring over all weather and large areas; it obtains the optimal variable combination through a variable selection strategy based on random forest importance ranking, ensuring the efficiency and robustness of the model; and it solves the problem of insufficient fitting ability of linear models through nonlinear machine learning algorithms, significantly improving estimation accuracy and stability. Attached Figure Description

[0025] Figure 1 This is a flowchart of the steps of the present invention;

[0026] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0027] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see the appendix Figure 1 -Appendix Figure 2 The present invention provides an embodiment of a method for estimating crop straw coverage based on fusion of optical spectral index and SAR remote sensing imagery, comprising the following steps: Step 1, data acquisition; Step 2, data preprocessing; Step 3, feature variable extraction; Step 4, variable screening and importance ranking; Step 5, collaborative inversion model construction; Step 6, regional straw coverage mapping.

[0029] In step one above, optical remote sensing images and SAR remote sensing images of the target monitoring area after crop harvest are acquired simultaneously. The preferred optical remote sensing image is Sentinel-2 MSI image, covering the visible, near-infrared, and short-wave infrared bands. The preferred SAR remote sensing image is Sentinel-1 satellite image data, in the interferometric wide-swath mode, with GRD product level and VV and VH dual polarization. VV is vertical transmission and vertical reception, and VH is vertical transmission and horizontal reception. Straw coverage data of sampling points in the target monitoring area are collected, and the center latitude and longitude coordinates of the sampling points are recorded to construct a ground-based measured dataset. The straw coverage data is calculated using either the transect method or the photographic method.

[0030] In step two above, the SAR remote sensing images acquired in step one undergo orbit correction, thermal noise removal, radiometric calibration, multi-view processing, filtering, terrain correction, decibel reduction, and resampling. Atmospheric correction is performed on the optical remote sensing images, converting the top-of-atmosphere reflectance to surface reflectance, and band resampling is performed to ensure consistency in spatial resolution and projection coordinate system between the optical and SAR images. Specifically, orbit correction uses orbit ephemeris files to correct satellite orbit information; thermal noise removal eliminates thermal noise generated by the radar system; radiometric calibration converts radar image pixel values ​​to radar backscattering coefficients; multi-view processing suppresses speckle noise; filtering removes speckle noise while preserving edge information, preferably using a Refined Lee filter for adaptive filtering; terrain correction uses a digital elevation model for range-Doppler terrain correction to eliminate geometric distortion caused by terrain undulations; decibel reduction converts linear unit backscattering coefficients to logarithmic units; and resampling resamples the SAR remote sensing images to a uniform spatial resolution.

[0031] In step three above, spectral indices sensitive to straw are extracted based on the preprocessed optical remote sensing image, and the VV polarization backscattering coefficient and VH polarization backscattering coefficient of the corresponding sampling point are extracted based on the preprocessed SAR remote sensing image, thus completing the extraction of multi-source feature variables. The spectral indices include NDI5, NDI7, NDSVI, SRNDI, NDRI, NDTI, and STI, preferably including NDTI and STI. The calculation formula is as follows:

[0032]

[0033]

[0034] in, The normalized differential tillage index. For simple farming index, It is the first band of shortwave infrared, with a center wavelength of 1610 nm. It is the second band of shortwave infrared, with a center wavelength of 2190nm;

[0035] In step four above, Pearson correlation coefficient analysis is used to remove feature variables that are weakly correlated with straw coverage, and the random forest algorithm is used to rank the variables by importance in order to select the optimal combination of feature variables. The optimal combination of feature variables selected is STI, NDTI, VH polarization backscattering coefficient and VV polarization backscattering coefficient.

[0036] In step five above, the optimal combination of feature variables selected in step four is used as the input variable, and the measured straw coverage on the ground is used as the output variable. A nonlinear collaborative estimation model for crop straw coverage is constructed and trained using either a backpropagation neural network (BP neural network) or a random forest. The prediction error is minimized by optimizing the model parameters. After training, the optimal model is selected by evaluating its accuracy. Specifically, the BP neural network model is constructed as follows: the number of input layer nodes corresponds to the number of feature variables; the hidden layer is processed using a transfer function; the output layer outputs the predicted coverage value; and the weights are adjusted using a backpropagation algorithm to minimize the prediction error. The random forest model is constructed as follows: multiple decision trees are constructed; a training set is generated using Bootstrap sampling; decision trees are grown by node splitting; and the predicted value is obtained by averaging the regression results of multiple trees.

[0037] In step six above, the optimal model selected in step five is used to perform pixel-by-pixel calculations on the remote sensing images of the entire monitoring area to generate a spatial distribution map of crop straw coverage at the regional scale, and coverage classification is completed according to a preset threshold.

[0038] Experimental Example 1:

[0039] The method for estimating crop straw coverage based on the fusion of optical spectral index and SAR remote sensing imagery includes the following steps: Step 1, data acquisition; Step 2, data preprocessing; Step 3, feature variable extraction; Step 4, variable screening and importance ranking; Step 5, collaborative inversion model construction; Step 6, regional straw coverage mapping.

[0040] In step one above, the corn-growing area in the southern part of the Songnen Plain in central Jilin Province was selected as the experimental area. The experiment was conducted during the bare soil period after the autumn corn harvest. Sentinel-2 MSI Level-1C imagery on October 20, 2020, was acquired as an optical remote sensing image with cloud cover below 10%. Sentinel-1A satellite imagery on November 4, 2020, was acquired as a SAR remote sensing imagery in interferometric wide-swath mode, GRD product level, and VV and VH dual polarization. From October 20 to October 30, 2020, 90 sampling points were set up in the experimental area. The corn stalk coverage at each sampling point was measured using the transect method. The transect was 40m long and stretched along the diagonal. The number of intersection points of the stalks was recorded to calculate the coverage.

[0041] In step two above, the SAR remote sensing images acquired in step one undergo orbit correction, thermal noise removal, radiometric calibration, multi-look processing, filtering, terrain correction, decibel reduction, and resampling. The Sen2Cor method is used to perform atmospheric correction on the optical remote sensing images, converting the top-of-atmosphere reflectance to surface reflectance, and resampling each band to a 20m resolution. Specifically, orbit correction uses orbital ephemeris files to correct satellite orbit information; thermal noise removal eliminates thermal noise generated by the radar system; radiometric calibration converts radar image pixel values ​​to radar backscattering coefficients; multi-look processing suppresses speckle noise; filtering uses a Refined Lee filter for adaptive filtering to remove speckle noise while preserving edge information, with a 7×7 window size; terrain correction uses a digital elevation model for range-Doppler terrain correction to eliminate geometric distortion caused by terrain undulations; decibel reduction converts linear unit backscattering coefficients to logarithmic units; and resampling resamples the SAR remote sensing images to a spatial resolution of 20m.

[0042] In step three above, spectral indices sensitive to straw are extracted based on the preprocessed optical remote sensing image, and the VV polarization backscattering coefficient and VH polarization backscattering coefficient of the corresponding sampling point are extracted based on the preprocessed SAR remote sensing image, thus completing the extraction of multi-source feature variables; the spectral indices include NDTI and STI, and the calculation formula is as follows:

[0043]

[0044]

[0045] in, The normalized differential tillage index. For simple farming index, It is the first band of shortwave infrared, with a center wavelength of 1610 nm. It is the second band of shortwave infrared, with a center wavelength of 2190nm;

[0046] In step four above, Pearson correlation coefficient analysis is used to remove feature variables that are weakly correlated with straw coverage, and the random forest algorithm is used to rank the variables by importance in order to select the optimal combination of feature variables. The optimal combination of feature variables selected is STI, NDTI, VH polarization backscattering coefficient and VV polarization backscattering coefficient.

[0047] In step five above, the optimal combination of feature variables selected in step four is used as the input variable, and the measured straw coverage on the ground is used as the output variable. A BP neural network is used to construct and train a nonlinear collaborative estimation model for crop straw coverage. The network structure is as follows: the number of nodes in the input layer is 4, the number of nodes in the hidden layer is set according to an empirical formula, and the number of nodes in the output layer is 1. The parameter settings are as follows: the transfer function of the hidden layer is the Sigmoid function, the transfer function of the output layer is the Linear function, the training algorithm is the Levenberg-Marquardt algorithm, the number of training iterations is set to 1000, and the learning rate is 0.01.

[0048] In step six above, the model trained in step five is used to perform pixel-by-pixel calculations on the remote sensing images of the entire experimental area to generate a spatial distribution map of crop straw coverage at the regional scale, and coverage classification is completed according to a preset threshold.

[0049] Experimental Example 2:

[0050] Referring to Experiment Example 1, only the BP neural network in step five is replaced with a random forest. The random forest algorithm is set as follows: the training set is generated based on the Bootstrap method, the number of decision trees is set to 1000, the number of node split variables is set to 2, the minimum number of samples per node is set to 5, and the output is the average value of the regression results of all decision trees as the final straw coverage prediction value.

[0051] Experimental Example 3:

[0052] To verify the superiority of this invention, the following comparative experiment was conducted: 90 ground-based measured samples were randomly divided into 60 modeling sets and 30 validation sets. Control group A used only the spectral index of Sentinel-2 optical images as input, and modeled using MLSR, RF, and BPNN respectively. Control group B used only the backscattering coefficient of Sentinel-1 radar images as input, and modeled using MLSR, RF, and BPNN respectively. The method proposed in this invention was used as the experimental group, using both Sentinel-2 optical index and Sentinel-1 radar backscattering coefficient as input, and modeling using MLSR, RF, and BPNN respectively. The coefficient of determination and root mean square error (RMSE) were used as evaluation indicators of model accuracy. The closer the coefficient of determination is to 1 and the smaller the RMSE, the higher the model accuracy. The models in each group performed well on the validation set. The accuracy statistics are shown in the table below. As can be seen from the table, the determination coefficient of control group B is only around 0.58, and the root mean square error is close to 10%, indicating that it is severely affected by soil moisture and roughness, making independent high-precision monitoring impossible. The accuracy of the experimental group is much higher than that of control group B, proving that the addition of optical information effectively constrains the uncertainty in radar signal interpretation. Although the determination coefficient of control group A reaches around 0.80, the accuracy of the experimental group after adding radar data is improved regardless of whether MLSR, RF, or BPNN algorithms are used. Specifically, the determination coefficient of the BPNN model in the experimental group is 0.836, and the root mean square error is 5.92%, proving that the texture and physical structure information provided by radar data can effectively correct background noise in optical inversion. Within the experimental group, BPNN and RF are superior to MLSR, with the BPNN model exhibiting the highest accuracy.

[0053]

[0054] Based on the above, the advantages of this invention are as follows: When used, it fully combines the complementary characteristics of optical remote sensing's sensitivity to straw biochemical components and SAR remote sensing's all-weather, 24 / 7 observation capabilities. Relying on a variable optimization strategy based on random forest importance ranking and a model construction method based on nonlinear machine learning algorithms such as BP neural networks and random forests, it effectively eliminates environmental interference such as soil background, moisture, and surface roughness, significantly improving estimation accuracy. The model's coefficient of determination can reach over 0.8, and the root mean square error is reduced to within 6%, far superior to the inversion effect of a single data source. Simultaneously, it overcomes the shortcomings of optical remote sensing, such as susceptibility to cloud and rain weather and data gaps, as well as the difficulties in SAR remote sensing interpretation and the low accuracy of standalone inversion. It achieves continuous, stable, and high-precision monitoring of straw coverage at the regional scale in complex farmland environments. Furthermore, this invention eliminates the need for extensive manual ground measurements, significantly reducing monitoring costs and time consumption. It can provide accurate and reliable data support for conservation tillage implementation assessment, agricultural subsidy policy formulation, black soil protection, and farmland ecological governance, demonstrating significant practicality and promotional value.

[0055] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for estimating crop straw coverage based on fused optical spectral index and SAR remote sensing imagery, comprising the following steps: Step 1: Data acquisition; Step 2: Data preprocessing; Step 3: Feature variable extraction; Step 4: Variable selection and importance ranking; Step 5: Collaborative inversion model construction; Step 6: Regional straw coverage mapping; Its key feature is: In step one above, optical remote sensing images and SAR remote sensing images of the target monitoring area after crop harvest are acquired simultaneously, straw coverage data of sampling points in the target monitoring area are collected, and the center latitude and longitude coordinates of the sampling points are recorded to construct a ground-based measured dataset. In step two above, the SAR remote sensing images acquired in step one are subjected to orbit correction, thermal noise removal, radiometric calibration, multi-view processing, filtering, terrain correction, decibel reduction, and resampling; atmospheric correction is performed on the optical remote sensing images, converting the top atmospheric reflectance to the surface reflectance, and band resampling is performed to ensure that the spatial resolution and projection coordinate system of the optical remote sensing images are consistent with those of the SAR remote sensing images. In step three above, spectral indices sensitive to straw are extracted based on the preprocessed optical remote sensing images, and VV polarization backscattering coefficients and VH polarization backscattering coefficients of the corresponding sampling points are extracted based on the preprocessed SAR remote sensing images, thus completing the extraction of multi-source feature variables. In step four above, correlation analysis is used to remove feature variables that are weakly correlated with straw coverage, and then the variables are sorted according to their importance to select the optimal combination of feature variables. In step five above, the optimal combination of feature variables selected in step four is used as the input variable, and the measured straw coverage on the ground is used as the output variable. The nonlinear collaborative estimation model of crop straw coverage is constructed and trained using machine learning algorithms. The prediction error is minimized by optimizing the model parameters. After the model training is completed, the optimal model is selected by evaluating the model accuracy. In step six above, the optimal model selected in step five is used to perform pixel-by-pixel calculations on the remote sensing images of the entire monitoring area to generate a spatial distribution map of crop straw coverage at the regional scale, and coverage classification is completed according to a preset threshold.

2. The method for estimating crop straw coverage based on fused optical spectral index and SAR remote sensing imagery according to claim 1, characterized in that: In step one, the preferred optical remote sensing image is Sentinel-2 MSI image, which covers the visible light, near-infrared and short-wave infrared bands; the preferred SAR remote sensing image is Sentinel-1 satellite image data, which is in interferometric wide-swath mode, GRD product level, and VV and VH dual polarization, where VV is vertical transmission and vertical reception, and VH is vertical transmission and horizontal reception.

3. The method for estimating crop straw coverage based on fused optical spectral index and SAR remote sensing imagery according to claim 1, characterized in that: In step one, the straw coverage data is calculated using either the transect method or the photographic method.

4. The method for estimating crop straw coverage based on fused optical spectral index and SAR remote sensing imagery according to claim 1, characterized in that: In step two, orbit correction involves correcting satellite orbit information using orbital ephemeris files; thermal noise removal eliminates thermal noise generated by the radar system; radiometric calibration converts radar image pixel values ​​into radar backscattering coefficients; multi-view processing performs multi-view processing on the image to suppress speckle noise; filtering removes speckle noise while preserving edge information, preferably using a Refined Lee filter for adaptive filtering; terrain correction uses a digital elevation model for range-Doppler terrain correction to eliminate geometric distortion caused by terrain undulations; and decibel conversion converts the backscattering coefficients in linear units to logarithmic units. Resampling is the process of resampling SAR remote sensing images to a uniform spatial resolution.

5. The method for estimating crop straw coverage based on fused optical spectral index and SAR remote sensing imagery according to claim 1, characterized in that: In step three, the spectral indices include NDI5, NDI7, NDSVI, SRNDI, NDRI, NDTI, and STI, preferably including NDTI and STI. The calculation formula is as follows: in, The normalized differential tillage index. For simple farming index, It is the first band of shortwave infrared, with a center wavelength of 1610 nm. It is the second band of shortwave infrared, with a center wavelength of 2190nm.

6. The method for estimating crop straw coverage based on fused optical spectral index and SAR remote sensing imagery according to claim 1, characterized in that: In step four, the correlation analysis uses the Pearson correlation coefficient, and the ranking is achieved using the variable importance assessment function of the random forest algorithm.

7. The method for estimating crop straw coverage based on fused optical spectral index and SAR remote sensing imagery according to claim 1, characterized in that: In step four, the optimal combination of feature variables selected is STI, NDTI, VH polarization backscattering coefficient and VV polarization backscattering coefficient.

8. The method for estimating crop straw coverage based on fused optical spectral index and SAR remote sensing imagery according to claim 1, characterized in that: In step five, the machine learning algorithm used is either a backpropagation neural network or a random forest.

9. The method for estimating crop straw coverage based on fused optical spectral index and SAR remote sensing imagery according to claim 8, characterized in that: The BP neural network model is constructed as follows: the number of input layer nodes corresponds to the number of feature variables, the hidden layer is processed by a transfer function, the output layer outputs the predicted coverage value, and the weights are adjusted by the backpropagation algorithm to minimize the prediction error.

10. The method for estimating crop straw coverage based on fused optical spectral index and SAR remote sensing imagery according to claim 8, characterized in that: The random forest construction model is as follows: construct multiple decision trees, use Bootstrap sampling to generate a training set, grow decision trees by splitting nodes, and finally obtain the predicted value by averaging the regression results of multiple trees.

Images