Soil attribute remote sensing inversion method integrating monthly synthetic image

Abstract

The invention discloses a soil attribute remote sensing inversion method integrated with monthly synthetic images, which comprises the following steps: S1, taking a soil sample collection year as a center, obtaining long-time sequence remote sensing images of continuous N years in a research area, and carrying out radiometric calibration and atmospheric correction preprocessing to obtain preprocessed remote sensing images; s2, for the preprocessed remote sensing image, removing pixels affected by cloud and cloud shadow to obtain a pure remote sensing image; s3, for the pure remote sensing image, synthesizing the remote sensing image in each month month by month to obtain the synthesized remote sensing image in each month; s4, integrating and utilizing the remote sensing images synthesized in each month, and constructing a soil attribute prediction model; and S5, inputting the integrated remote sensing data set of each month into the soil attribute prediction model to generate a soil attribute remote sensing prediction map. According to the method, remote sensing images of different months are integrated and applied, so that the problem of month selection is avoided, remote sensing information is increased, and the precision of soil attribute prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a method for remotely sensing and inverting soil properties by integrating monthly synthesized images. Background Art

[0002] In the field of soil science research, the accurate acquisition of soil properties is of great significance for aspects such as agricultural management, environmental monitoring, and land resource utilization. Traditional soil property measurements rely on field sampling and laboratory analysis. Although this method is accurate, it is time-consuming and laborious, and it is difficult to achieve large-scale real-time monitoring. Therefore, as a non-contact monitoring method, remote sensing technology has gradually been widely used in the inversion of soil properties.

[0003] However, remote sensing technology also faces many challenges in the inversion of soil properties. Among them, the selection of effective remote sensing image months is a key issue. Since the change of soil properties is affected by various factors, such as climate, vegetation cover, etc., the soil information reflected by remote sensing images in different months is different. Therefore, when constructing a soil property prediction model, how to select appropriate remote sensing image months has become a difficult problem. If the selection is improper, it may lead to a decrease in the prediction accuracy of the model and cannot accurately reflect the true situation of soil properties.

[0004] To overcome this difficult problem, researchers have tried to use various methods for processing. For example, some methods only select remote sensing images of specific months for inversion. Although this method is simple and easy to implement, it may ignore the soil information contained in other months, resulting in an incomplete inversion result. Another method is to try to merge remote sensing images of multiple months for processing in order to obtain richer soil information. However, this method often faces technical problems such as data fusion and information extraction in actual operation, and it is difficult to achieve high-precision inversion of soil properties. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to provide a method for remotely sensing and inverting soil properties by integrating monthly synthesized images. By synthesizing remote sensing images month by month, the information in remote sensing images of different months is fully utilized to construct a high-precision soil property prediction model.

[0006] Technical Solution: A method for remotely sensing and inverting soil properties by integrating monthly synthesized images includes the following steps: S1, centering on the year of soil sample collection, obtain long-term time series remote sensing images of the study area for N consecutive years, and perform preprocessing of radiometric calibration and atmospheric correction to obtain preprocessed remote sensing images; S2, for the preprocessed remote sensing images, remove the pixels affected by clouds and cloud shadows to obtain pure remote sensing images; S3. For the pure remote sensing images, synthesize the remote sensing images within each month month by month to obtain the synthesized remote sensing images for each month. The implementation steps are as follows: S31. For the obtained pure remote sensing images, divide them into sets of remote sensing images within different months according to months to obtain sets of remote sensing images for each month from January to December; S32. For each set of remote sensing images for each month, use the method of pixel-by-pixel synthesis and the median or mean method to synthesize the remote sensing images for each month month by month to obtain the synthesized remote sensing images for each month from January to December; S4. Integrate and utilize the synthesized remote sensing images for each month to construct a soil property prediction model; S5. Input the integrated remote sensing data sets for each month into the soil property prediction model to generate a remote sensing prediction map of soil properties.

[0007] Furthermore, the implementation steps for obtaining the preprocessed remote sensing images are as follows: S11. Select a remote sensing image source with a time resolution less than 28 days to ensure that remote sensing images can be obtained for each month within a year; S12. Taking the year of soil sample collection as the center, obtain the remote sensing images of the study area for continuous N years, where N≥3; S13. For the obtained remote sensing images, perform radiometric calibration processing according to the following formula to obtain radiance remote sensing data: L = DN*α + β where L is the image radiance value obtained after calibration, DN is the remote sensing image value obtained in step S12, and α and β are the coefficients for converting the image DN value to radiance respectively; S14. For the radiance remote sensing data, perform atmospheric correction processing using the FLAASH model to obtain the preprocessed remote sensing images.

[0008] Furthermore, the implementation steps for obtaining the pure remote sensing images are as follows: S21. For each scene of the preprocessed remote sensing images, use the remote sensing image quality control band to screen the remote sensing pixels affected by clouds and cloud shadows; S22. Perform masking processing on the pixels affected by clouds and cloud shadows to obtain pure remote sensing images that are not affected by clouds and cloud shadows.

[0009] Furthermore, the implementation steps for constructing the soil property prediction model are as follows: S41. For the synthesized remote sensing images for each month obtained, combine and integrate the data sets of remote sensing images for each month from January to December to obtain the integrated remote sensing data sets for each month, and obtain the target soil property data within the remote sensing inversion area; S42. Divide the obtained target soil attribute data into a training set and a validation set according to a certain ratio, where the number of soil attribute data in the training set is greater than that in the validation set. S43. Use the training set and a machine learning algorithm to construct a prediction model for the target soil attributes to obtain a soil attribute prediction model; use the validation set to evaluate the accuracy of the soil attribute prediction model.

[0010] Furthermore, adopt the method of cross-validation or independent validation, and use the coefficient of determination R 2 and the root mean square error RMSE as two indicators to evaluate the accuracy of the soil attribute prediction model, where the larger the R 2 value and the smaller the RMSE value indicate higher model accuracy.

[0011] Compared with the prior art, the remarkable effects of the present invention are as follows: According to the remotely sensed images obtained continuously for N years in the research area, the present invention synthesizes the remotely sensed images month by month, and then constructs a soil attribute prediction model using the remotely sensed images synthesized in all months, which can solve the problem of selecting the months of remotely sensed images when predicting soil attributes using long-term time series remotely sensed images; at the same time, due to the integrated application of remotely sensed images in different months, the soil information in the remotely sensed images in different months is fully utilized, increasing the remotely sensed information when the remotely sensed images predict soil attributes, improving the accuracy of soil attribute prediction, and promoting the practical application of the soil attribute remote sensing inversion technology. Description of the Drawings

[0012] Figure 1 is the flow chart of the present invention; Figure 2 is the remotely sensed image map synthesized month by month in the implementation case; Figure 3 is the spatial distribution map of soil sample collection points in the implementation case; Figure 4 is the statistical distribution map of soil organic matter in the implementation case; Figure 5 is the scatter plot of the soil organic matter prediction results obtained in the implementation case; Figure 6 is the spatial prediction map of soil organic matter obtained in the implementation case. Detailed Embodiments

[0013] The present invention will be further described in detail below in conjunction with the drawings of the specification and the detailed embodiments.

[0014] The present invention designs a remote sensing inversion method for soil attributes integrating monthly synthesized images for obtaining the spatial distribution of soil attributes. The flow chart is as Figure 1 shown, and specifically includes the following steps: Step 1: Select long-time series remote sensing images, perform radiometric calibration and atmospheric correction preprocessing to obtain preprocessed remote sensing images, and then proceed to Step 2.

[0015] In practical applications, the above Step 1 is specifically executed as Steps 11 to 14 below: Step 11: Select a remote sensing image source with a time resolution less than 28 days to ensure that remote sensing images can be obtained for each month of the year, and then proceed to Step 12; Step 12: Centering on the year of soil sample collection, obtain continuous N-year remote sensing images within the study area, where N ≥ 3, to ensure that no less than 3 scenes of remote sensing images can be obtained for different months. If the obtained remote sensing images have already completed radiometric calibration and atmospheric correction processing, directly proceed to Step 2; Step 13: For the obtained remote sensing images, perform radiometric calibration processing according to the following formula to obtain radiance remote sensing data: L = DN*α + β where DN is the remote sensing image value obtained in Step 12, L is the image radiance value obtained after calibration, and α and β are the coefficients for the conversion of the image DN value to radiance respectively, and then proceed to Step 14; Step 14: For the radiance remote sensing data obtained after radiometric calibration, use the FLAASH model to perform atmospheric correction processing to obtain preprocessed remote sensing images, and then proceed to Step 2.

[0016] Step 2: For the preprocessed remote sensing images, remove the pixels affected by clouds and cloud shadows to obtain pure remote sensing images, and then proceed to Step 3.

[0017] In practical applications, the above Step 2 is specifically executed as Steps 21 to 22 below: Step 21: For each scene of preprocessed remote sensing images, use the remote sensing image quality control band to screen the remote sensing pixels affected by clouds and cloud shadows, and then proceed to Step 22; Step 22: Perform masking processing on the screened pixels affected by clouds and cloud shadows to obtain pure remote sensing images that are not affected by clouds and cloud shadows, and then proceed to Step 3.

[0018] Step 3: For the pure remote sensing images, synthesize the remote sensing images within each month month by month to obtain the synthesized remote sensing images for each month, and then proceed to Step 4.

[0019] In practical applications, the above Step 3 is specifically executed as Steps 31 to 32 below.

[0020] Step 31: For the obtained pure remote sensing images, divide them into remote sensing image sets within different months according to months to obtain remote sensing image sets for each month from January to December, and then proceed to Step C2; Step 32: For the remote sensing image sets of each month, adopt the method of synthesizing pixel by pixel of the remote sensing images, and use the median or mean method to synthesize the remote sensing images of each month month by month, obtaining the synthesized remote sensing images of each month from January to December, and then proceed to Step 4.

[0021] Step 4: Integrate and utilize the synthesized remote sensing images of each month to construct a soil property prediction model, and then proceed to Step 5.

[0022] In practical applications, the above Step 4 is specifically executed as follows in Steps 41 to 43.

[0023] Step 41: For the synthesized remote sensing images of each month obtained, combine and integrate the data sets of the remote sensing images of each month from January to December to obtain the integrated remote sensing data sets of each month, and obtain the target soil property data within the remote sensing inversion area; Step 42: For the obtained target soil property data, divide it into a training set and a validation set according to a certain ratio, where the quantity requirement of the soil property data in the training set is greater than that in the validation set, and the soil properties may include soil organic matter, total soil nitrogen, soil texture, soil salinity, etc.; Step 43: Use machine learning algorithms such as Random Forest with the training set to construct a prediction model for the target soil property, and train the constructed prediction model to obtain a soil property prediction model, and then proceed to Step 5.

[0024] Step 5: Conduct accuracy evaluation on the constructed soil property prediction model to generate a remote sensing prediction map of soil properties.

[0025] In practical applications, the above Step 5 is specifically executed as follows in Steps 51 to 52.

[0026] Step 51: For the obtained soil property prediction model, adopt the method of cross-validation or independent validation, and use two indicators, the coefficient of determination R 2 and the root mean square error RMSE, to evaluate the model accuracy, where the larger the R 2 value and the smaller the RMSE value indicate higher model accuracy, and then proceed to Step 52; Step 52: Input the obtained integrated remote sensing data sets of each month into the constructed soil property prediction model to generate a remote sensing prediction map of soil properties.

[0027] Apply the above-designed remote sensing inversion method of soil properties integrating monthly composite images to practice. Taking Youyi County, Heilongjiang Province as the remote sensing inversion area, the farmland soil organic matter collected in 2021 as the inversion object, and the long-term Landsat-8 remote sensing images as the image source, the remote sensing inversion of farmland soil organic matter is taken as an example for further detailed description, but it is not a limitation to the present invention. The specific implementation is as follows.

[0028] Step A, select the long-term Landsat-8 remote sensing images, perform radiometric calibration and atmospheric correction preprocessing to obtain the preprocessed remote sensing images, and then enter Step B.

[0029] The above Step A is specifically as follows in the embodiment: Step A1, select Landsat-8 images with a temporal resolution of 16 days and a spatial resolution of 30m as the image source to ensure that remote sensing images can be obtained for each month of the year. The image bands used include: Band 1: 0.435 - 0.451 µm, Band 2: 0.452 - 0.512 µm, Band 3: 0.533 - 0.590 µm, Band 4: 0.636 - 0.673 µm, Band 5: 0.851 - 0.879 µm, Band 6: 1.566 - 1.651 µm, Band 7: 2.107 - 2.294 µm, Band 10: 10.60 - 11.19 µm; then enter Step A2; Step A2, the year of soil sample collection is 2021. Centered on 2021, obtain 446 remote sensing images of Youyi County continuously for 6 years from January 2018 to December 2023 to ensure that no less than 3 remote sensing images can be obtained for different months; since the obtained images have completed radiometric calibration and atmospheric correction processing, directly enter Step B; Step B, for the obtained 446 Landsat-8 remote sensing images, remove the pixels affected by clouds and cloud shadows to obtain pure remote sensing images, and then enter Step C.

[0030] The above Step B is specifically as follows in the embodiment: Step B1, for each Landsat-8 remote sensing image, use the quality control band attached to the remote sensing image to screen out the remote sensing pixels affected by clouds and cloud shadows, and then enter Step B2; Step B2, for each Landsat-8 remote sensing image, mask and remove the pixels affected by clouds and cloud shadows screened out to obtain pure remote sensing images not affected by clouds and cloud shadows, and then enter Step C.

[0031] Step C: For the pure remote sensing images, synthesize the remote sensing images within each month month by month to obtain the synthesized remote sensing images for each month, and then proceed to Step D.

[0032] The above Step C is specifically as follows in the embodiment: Step C1: For the obtained 446 scenes of pure remote sensing images, divide them into remote sensing image sets within different months according to months to obtain remote sensing image sets for each month from January to December. Each monthly remote sensing image set includes remote sensing images for six consecutive years; then proceed to Step C2; Step C2: For the remote sensing image sets for each month from January to December, use the method of synthesizing each image pixel by pixel, and use the median method to synthesize the remote sensing images for each month month by month to obtain the synthesized remote sensing images for each month from January to December (as Figure 2 shown), and then proceed to Step D.

[0033] Step D: Integrate and utilize the synthesized remote sensing images for each month to construct a soil organic matter prediction model, and then proceed to Step E.

[0034] The above Step D is specifically as follows in the embodiment: Step D1: For the synthesized remote sensing images for each month obtained, combine and integrate the data sets of remote sensing images for each month from January to December to obtain the integrated remote sensing data sets for each month, and obtain the soil organic matter data within the remote sensing inversion area, and then proceed to Step D2; Step D2: For the soil organic matter data, divide the obtained 109 soil organic matter data (as Figure 3 shown) into a training set and a validation set according to a ratio of approximately 70% and 30%. Among them, the number of samples in the training set is 72, and the amount of sample data in the validation set is 37. The obtained 109 soil organic matter data approximately conforms to a normal distribution (as Figure 4 shown), and the average value, minimum value, and maximum value are respectively: 2.075%, 4.105%, and 6.845%, and then proceed to Step D3; Step D3: Use the training set and the Random Forest algorithm to construct a soil organic matter prediction model. In the Random Forest algorithm, the number of trees is set to: 500, the variable division number is set to: the square root of the number of variables, the minimum number of leaf nodes is set to: 1, the maximum number of leaf nodes is not limited. Use the training set to train the constructed prediction model to obtain the soil organic matter prediction model, and then proceed to Step E.

[0035] Step E: Conduct accuracy evaluation on the constructed soil organic matter prediction model to generate a remote sensing prediction map of soil organic matter.

[0036] The above Step E is specifically as follows in the embodiment: Step E1. For the obtained soil organic matter prediction model, calculate the coefficient of determination R using the validation set 2 and the root mean square error RMSE, where R 2 is 0.68 and RMSE is 0.69% (as Figure 5 shown), and then proceed to Step E2; Step E2. Input the obtained integrated remote sensing data sets for each month into the soil organic matter prediction model to generate a remote sensing prediction map of soil organic matter (as Figure 6 shown).

[0037] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the gist of the present invention within the scope of knowledge possessed by those of ordinary skill in the art.

Claims

1. A soil property remote sensing inversion method integrating monthly synthetic images, characterized in that: The steps include: S1, taking the soil sample collection year as the center, obtain the long-term remote sensing image series of N consecutive years in the study area, perform radiation calibration and atmospheric correction preprocessing, and obtain the preprocessed remote sensing image; S2, for the pre-processed remote sensing image, remove the pixels affected by clouds and cloud shadows to obtain a pure remote sensing image; S3, for pure remote sensing images, synthesize the remote sensing images of each month month by month to obtain the synthesized remote sensing images of each month; the implementation steps are as follows: S31, dividing the obtained pure remote sensing images into remote sensing image sets in different months according to the months, and obtaining remote sensing image sets for each month from January to December; S32, for the remote sensing image set of each month, adopting a method of synthesizing each image pixel by pixel, using a median or mean method, synthesizing the remote sensing images of each month month by month, and obtaining synthetic remote sensing images of each month from January to December; S4, integrates and utilizes the remote sensing images synthesized in each month to build a soil property prediction model; S5, input the integrated remote sensing data set of each month into the soil property prediction model to generate a soil property remote sensing prediction map.

2. The soil property remote sensing inversion method integrating monthly synthetic images according to claim 1 is characterized in that: The steps to obtain the preprocessed remote sensing image are as follows: S11, select remote sensing image sources with a temporal resolution of less than 28 days to ensure that remote sensing images can be obtained in every month of the year; S12, taking the soil sample collection year as the center, obtain remote sensing images of N consecutive years in the study area, where N ≥ 3; S13, for the acquired remote sensing image, perform radiation calibration processing according to the following formula to obtain radiance remote sensing data: L = DN*α+ β Wherein, L is the image radiance value obtained after calibration, DN is the remote sensing image value obtained in step S12, and α and β are the conversion coefficients of image DN value and radiance respectively; S14, for the radiance remote sensing data, the atmospheric correction processing is performed using the FLAASH model to obtain a preprocessed remote sensing image.

3. The soil property remote sensing inversion method integrating monthly synthetic images according to claim 1 is characterized in that: The steps to obtain a pure remote sensing image are as follows: S21, for each pre-processed remote sensing image, use the remote sensing image quality control band to screen the remote sensing pixels affected by clouds and cloud shadows; S22, performing mask processing on pixels affected by clouds and cloud shadows to obtain a pure remote sensing image that is not affected by clouds and cloud shadows.

4. The soil property remote sensing inversion method integrating monthly synthetic images according to claim 1 is characterized in that: The implementation steps for building a soil property prediction model are as follows: S41, combining and integrating the data sets of the remote sensing images of each month from January to December for the obtained synthetic remote sensing images, obtaining the integrated remote sensing data sets of each month, and obtaining the target soil property data in the remote sensing inversion area; S42, dividing the acquired target soil property data into a training set and a validation set according to a certain ratio, wherein the amount of soil property data in the training set is greater than the amount of soil property data in the validation set; S43, using the training set and the machine learning algorithm to construct a prediction model of the target soil properties, and obtain the soil property prediction model; using the validation set to evaluate the accuracy of the soil property prediction model.

5. The soil property remote sensing inversion method integrating monthly synthetic images according to claim 4 is characterized in that: Use cross-validation or independent validation methods and use the determination coefficient R 2 The accuracy of soil property prediction model is evaluated by two indicators: R 2 The larger the value and the smaller the RMSE value, the higher the model accuracy.

Images