Single-cropping and intercropping corn remote sensing mapping method based on phenological assistance and machine learning

Abstract

The invention relates to the technical field of remote sensing mapping, in particular to a single-cropping and intercropping corn remote sensing mapping method based on phenological assistance and machine learning, which comprises the following steps: inputting extraction results of a plurality of spectral bands and vegetation indexes of a farmland area in a research area into a trained classification model to obtain a preliminary classification result map; calculating a classification precision index for verification by using the constructed verification data set; and area correction is carried out in combination with regional agricultural statistical data, and final monoculture and intercropping corn spatial distribution diagrams and area estimation results are output. Through organic combination of phenology-assisted automatic sample generation and machine learning classification, the core technical problem of large-scale monoplot and intercropping corn mapping is solved, and the method has the characteristics of high precision, high automation and high adaptability, can be popularized and applied to complex planting system mapping of other small agricultural areas in the world, and has a wide application prospect. And a technical support is provided for grain safety and agricultural sustainable development.

Description

Technical Field

[0001] This invention relates to the field of remote sensing mapping technology, specifically to a remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning. Background Technology

[0002] Maize is a vital food crop globally, with both monoculture and intercropping with legumes playing significant roles in agricultural production. Intercropping, in particular, helps maintain soil fertility and ensure food security in small-scale farming regions like Kenya. However, existing large-scale remote sensing mapping methods struggle to accurately distinguish between monoculture and intercropped maize due to issues such as spectral confusion, significant crop phenological variability, and difficulties in obtaining samples.

[0003] Traditional crop mapping methods are mostly designed for large-scale monoculture agriculture, and face challenges in small-scale farming systems such as fragmented plots, diverse management practices, and inconsistent phenological periods. Manual sampling methods are time-consuming, labor-intensive, and prone to errors, while existing automated sampling frameworks lack adaptability to intercropping systems. This results in low accuracy and poor applicability in large-scale monoculture and intercropping maize mapping, failing to meet the monitoring needs of sustainable agricultural development. Therefore, existing technologies have significant shortcomings. Summary of the Invention

[0004] The purpose of this invention is to provide a remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning, so as to solve the problems mentioned in the background art.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning, the method comprising the following steps: S1. Acquire multispectral remote sensing images of the study area, perform cloud removal, temporal synthesis and missing value interpolation on the acquired multispectral remote sensing images, and extract multiple spectral bands and vegetation indices. S2. Multi-source land use data are fused to generate farmland masks, eliminating pixel interference from non-farmland areas in the study area, and extracting multiple spectral bands and vegetation indices of farmland areas in the study area. S3. Through field surveys, the center coordinates and growth stage information of monoculture maize, intercropped maize, and non-maize crops were recorded. Combined with high-resolution Google Earth imagery to supplement the samples, training and validation datasets were constructed. Separability analysis was used to determine the optimal time window for the vegetation pigment variation index, and thresholds were set based on the statistical distribution of the vegetation pigment variation index to filter maize pixels. Separability analysis was also used to determine the optimal time window for the moisture and greenness ratio index, distinguishing between monoculture and intercropped maize pixels. Different thresholds were set based on the statistical distribution of the moisture and greenness ratio indices corresponding to monoculture and intercropped maize to identify intercropped and monoculture maize samples respectively. S4. Combine the monoculture maize samples, intercropping maize samples and non-maize crop samples with a preset ratio obtained in step S3 into a training set, select multiple spectral bands and vegetation indices as input features, and train the classification model using the random forest algorithm. S5. Input the extracted results of multiple spectral bands and vegetation indices of farmland areas in the study area into the trained classification model to obtain a preliminary classification result map; use the constructed validation dataset to calculate the classification accuracy index for validation; and combine regional agricultural statistics to perform area correction, and output the final spatial distribution map of monoculture and intercropping maize and the area estimation results.

[0006] Furthermore, the multiple spectral bands extracted in S1 include reflectance values ​​of the green band, red band, red-edge band 1, short-wave infrared band 1, and near-infrared band. The vegetation index includes the soil-optimized vegetation index, the converted chlorophyll absorption and reflectance index, and the normalized vegetation index. The formula for calculating the optimized soil-adjusted vegetation index is: OSAVI = 1.16 × (NIR - R) / (NIR + R + 0.16), where OSAVI represents the optimized soil-adjusted vegetation index; NIR represents the reflectance value of the near-infrared band at the corresponding time point; and R represents the reflectance value of the red band at the corresponding time point. The formula for calculating the converted chlorophyll absorption reflectance index is: TCARI=3×[(RE1-R)-0.2×(RE1-G)×RE1 / R], where TCARI represents the converted chlorophyll absorption reflectance index; RE1 represents the reflectance value of the red-edge band 1 at the corresponding time point; and G represents the reflectance value of the green band at the corresponding time point. The formula for calculating the Normalized Difference Vegetation Index (NDVI) is: NDVI = (NIR - R) / (NIR + R), where NDVI represents the Normalized Difference Vegetation Index.

[0007] This invention acquires surface reflectance data from Sentinel-2 in the study area using Google Earth engine, removes cloud-covered pixels using scene classification markers and cloud probability products, synthesizes time-series images over 10 days, and fills in missing values ​​using time-linear interpolation. Furthermore, by extracting multiple spectral bands such as green, red, red-edge 1, near-infrared, and shortwave infrared 1, it calculates and optimizes multiple vegetation indices, including soil-adjusted vegetation index, converted chlorophyll absorption reflectance index, and normalized vegetation index, forming a classification feature set. This provides data basis for effectively distinguishing between monoculture and intercropped maize.

[0008] Furthermore, the farmland mask is generated by fusing Esri land cover data, ESA world cover data, and Google Dynamic World data. The fused results are then subjected to intersection or union operations, followed by visual verification using high-resolution imagery and manual correction to remove non-farmland plots. This invention fuses multi-source 10-meter resolution land use data and, through union operations and high-resolution image verification, removes non-farmland pixels such as forests and built-up areas, reducing classification errors.

[0009] Furthermore, the vegetation pigment variation index is obtained by multiplying the vegetation and chlorophyll variation index by the anthocyanin dynamic variation index; the optimal time window corresponding to the vegetation and chlorophyll variation index and the anthocyanin dynamic variation index is determined by the separability analysis, and a threshold is set based on the statistical distribution of the vegetation pigment variation index of non-corn crops to screen out corn pixels. The vegetation and chlorophyll variation index is calculated by accumulating the absolute values ​​of the differences between the optimized soil-adjusted vegetation index and the converted chlorophyll absorption and reflection index at each time point after normalization during the key phenological period of maize. The optimized soil-adjusted vegetation index and the converted chlorophyll absorption and reflection index are normalized in the same way, and the normalized values ​​of the optimized soil-adjusted vegetation index and the converted chlorophyll absorption and reflection index are both in the range of [0, 1]. The normalized value of the optimized soil-adjusted vegetation index is equal to the difference between the optimized soil-adjusted vegetation index at the corresponding time point during the key phenological period of maize and the minimum value of the optimized soil-adjusted vegetation index at each time point during the key phenological period of maize, divided by the difference between the maximum and minimum optimized soil-adjusted vegetation index at each time point during the key phenological period of maize. The normalized value of the converted chlorophyll absorption reflectance index is equal to the difference between the converted chlorophyll absorption reflectance index at the corresponding time point during the key phenological period of maize and the minimum value of the converted chlorophyll absorption reflectance index at each time point during the key phenological period of maize, divided by the difference between the maximum and minimum converted chlorophyll absorption reflectance index at each time point during the key phenological period of maize. The anthocyanin dynamic variation index is calculated by accumulating the absolute values ​​of the first difference of the anthocyanin reflectance index at each time point during the key phenological period of maize.

[0010] Furthermore, through separability analysis, the optimal time window for the vegetation pigment variation index was determined to be 10 to 30 days before corn tasseling, and the optimal time window for the anthocyanin dynamic variation index was 40 to 60 days before corn tasseling. The threshold for the vegetation pigment variation index was set to 0.05, which is the 95th percentile of the distribution of vegetation pigment variation index values ​​for non-corn crops.

[0011] Furthermore, the moisture to greenness ratio index is the ratio of the mean of the shortwave infrared 1 band to the mean of the normalized vegetation index. The optimal time window for the normalized vegetation index and the shortwave infrared 1 band is determined by the separability analysis, and different thresholds are set based on the statistical distribution of the moisture to greenness ratio index corresponding to monoculture and intercropping maize, so as to identify intercropping maize samples and monoculture maize samples respectively. Separability analysis determined that the optimal time window for the Normalized Difference Vegetation Index (NDI) was 40 to 60 days before maize tasseling, and the optimal time window for shortwave infrared 1 band was 0 to 20 days after maize tasseling. A moisture-to-greenness ratio (MDR) greater than 2.0 was set as the threshold for identifying intercropped maize, corresponding to the 95th percentile of the WGR value distribution of monoculture maize; a MDR less than 1.0 was set as the threshold for identifying monoculture maize, corresponding to the 5th percentile of the WGR value distribution of intercropped maize.

[0012] Furthermore, the separability analysis uses the separability index to quantitatively assess the distinguishability of spectral or index features of two types of crops within a specific time window. The calculation formula is the ratio of the mean difference of the corresponding feature values ​​of the two types of crops to the sum of the standard deviations of the corresponding feature values ​​of the two types of crops by 1.96 times. The larger the value of the separability index, the better the separability. The time window with the largest value of the separability index is selected as the optimal time window.

[0013] Furthermore, in the process of training the classification model using the random forest algorithm in S4, the parameters of the random forest algorithm model are set as follows: the number of decision trees is 200, and the number of features randomly selected when each decision tree grows is the square root of the total number of input features; the input features include the reflectance values ​​of the green band, red band, red-edge 1 band, near-infrared band, and short-wave infrared 1 band, as well as the normalized vegetation index and the converted chlorophyll absorption reflectance index.

[0014] This invention employs the random forest algorithm to construct a classification model. This algorithm reduces the risk of overfitting by integrating multiple decision trees and is suitable for high-dimensional remote sensing feature classification. During training, the model is trained using automatically generated monoculture and intercropping maize samples, as well as some non-maize crop samples, to address the phenological differences between long and short rainy seasons. The model parameters are optimized to balance classification accuracy and computational efficiency.

[0015] Furthermore, the classification accuracy indicators include overall accuracy, producer accuracy, user accuracy, Kappa coefficient, and F1 score; the area correction adopts the linear scaling method to match and adjust the total national or regional area estimated by the model with the official statistical area of ​​the corresponding region; the classification model outputs a spatial distribution map of monoculture and intercropping maize with a resolution of 10 meters.

[0016] Compared with the prior art, the beneficial effects achieved by the present invention are: (1) This invention solves the core technical problem of large-scale monoculture and intercropping maize mapping by combining phenological-assisted automated sample generation with machine learning classification. It has the characteristics of high precision, high automation and strong adaptability. It can be extended to the mapping of complex planting systems in other small-scale agricultural areas around the world, providing technical support for food security and sustainable agricultural development. (2) This invention achieves automated sample generation through stratified phenological index, which greatly reduces the workload of manual sampling, reduces dependence on ground data, has a high degree of automation, and is suitable for small-scale agricultural areas with scarce data. (3) This invention determines the optimal time window through separability analysis and combines it with the relative phenology calculation method, which can adapt to phenological variations in different regions and can be extended to complex agricultural systems worldwide; (4) The distribution map constructed by this invention can accurately capture the spatial details and seasonal dynamics of monoculture and intercropping maize, providing core data support for agricultural policy formulation, yield estimation, climate change adaptation and monitoring of the zero hunger sustainable development goal. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning, as described in this invention. Figure 2 This is a map showing the distribution of monoculture and intercropping maize during the long rainy season of 2023, based on the remote sensing mapping method for monoculture and intercropping maize using phenological assistance and machine learning, as described in this invention. Detailed Implementation

[0018] 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.

[0019] Please see Figures 1-2 This example provides a remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning. The method includes the following steps: S1. Acquire multispectral remote sensing images of the study area, perform cloud removal, temporal synthesis and missing value interpolation on the acquired multispectral remote sensing images, and extract multiple spectral bands and vegetation indices. The multiple spectral bands extracted in S1 include the reflectance values ​​of the green band, red band, red-edge band 1, short-wave infrared band 1, and near-infrared band. The vegetation index includes the soil-optimized vegetation index, the converted chlorophyll absorption and reflectance index, and the normalized vegetation index. The formula for calculating the optimized soil-adjusted vegetation index is: OSAVI = 1.16 × (NIR - R) / (NIR + R + 0.16), where OSAVI represents the optimized soil-adjusted vegetation index; NIR represents the reflectance value of the near-infrared band at the corresponding time point; and R represents the reflectance value of the red band at the corresponding time point. The formula for calculating the converted chlorophyll absorption reflectance index is: TCARI=3×[(RE1-R)-0.2×(RE1-G)×RE1 / R], where TCARI represents the converted chlorophyll absorption reflectance index; RE1 represents the reflectance value of the red-edge band 1 at the corresponding time point; and G represents the reflectance value of the green band at the corresponding time point. The formula for calculating the Normalized Difference Vegetation Index (NDVI) is: NDVI = (NIR - R) / (NIR + R), where NDVI represents the Normalized Difference Vegetation Index.

[0020] S2. Multi-source land use data are fused to generate farmland masks, eliminating pixel interference from non-farmland areas in the study area, and extracting multiple spectral bands and vegetation indices of farmland areas in the study area. The farmland mask is generated by fusing Esri land cover data, ESA world cover data, and Google Dynamic World data. The fused results are then subjected to intersection or union operations, followed by visual verification using high-resolution imagery and manual correction to remove non-farmland plots. This invention fuses multi-source 10-meter resolution land use data and, through union operations and high-resolution image verification, removes non-farmland pixels such as forests and built-up areas, reducing classification errors.

[0021] S3. Through field surveys, the center coordinates and growth stage information of monoculture maize, intercropped maize, and non-maize crops were recorded. Combined with high-resolution Google Earth imagery to supplement the samples, training and validation datasets were constructed. Separability analysis was used to determine the optimal time window for the vegetation pigment variation index, and thresholds were set based on the statistical distribution of the vegetation pigment variation index to filter maize pixels. Separability analysis was also used to determine the optimal time window for the moisture and greenness ratio index, distinguishing between monoculture and intercropped maize pixels. Different thresholds were set based on the statistical distribution of the moisture and greenness ratio indices corresponding to monoculture and intercropped maize to identify intercropped and monoculture maize samples respectively. The vegetation pigment variation index is obtained by multiplying the vegetation and chlorophyll variation index by the anthocyanin dynamic variation index. Separability analysis is used to determine the optimal time windows corresponding to the vegetation and chlorophyll variation indices and the anthocyanin dynamic variation index, respectively. A threshold is set based on the statistical distribution of the vegetation pigment variation index of non-corn crops to filter out corn pixels. Separability analysis determines the optimal time window for the vegetation pigment variation index to be 10 to 30 days before corn tasseling, and the optimal time window for the anthocyanin dynamic variation index to be 40 to 60 days before corn tasseling. The threshold for the vegetation pigment variation index is set to 0.05, which is the 95th percentile of the distribution of vegetation pigment variation index values ​​for non-corn crops.

[0022] The vegetation and chlorophyll variation index is calculated by accumulating the absolute values ​​of the differences between the optimized soil-adjusted vegetation index and the converted chlorophyll absorption and reflection index at each time point after normalization during the key phenological period of maize. The optimized soil-adjusted vegetation index and the converted chlorophyll absorption and reflection index are normalized in the same way, and the normalized values ​​of the optimized soil-adjusted vegetation index and the converted chlorophyll absorption and reflection index are both in the range of [0, 1]. The normalized value of the optimized soil-adjusted vegetation index is equal to the difference between the optimized soil-adjusted vegetation index at the corresponding time point during the key phenological period of maize and the minimum value of the optimized soil-adjusted vegetation index at each time point during the key phenological period of maize, divided by the difference between the maximum and minimum optimized soil-adjusted vegetation index at each time point during the key phenological period of maize. The normalized value of the converted chlorophyll absorption reflectance index is equal to the difference between the converted chlorophyll absorption reflectance index at the corresponding time point during the key phenological period of maize and the minimum value of the converted chlorophyll absorption reflectance index at each time point during the key phenological period of maize, divided by the difference between the maximum and minimum converted chlorophyll absorption reflectance index at each time point during the key phenological period of maize. The anthocyanin dynamic variation index is calculated by accumulating the absolute values ​​of the first difference of the anthocyanin reflectance index at each time point during the key phenological period of maize.

[0023] The moisture-to-greenness ratio index is the ratio of the mean of the shortwave infrared 1 band to the mean of the normalized vegetation index. The optimal time window for the normalized vegetation index and the shortwave infrared 1 band is determined by the separability analysis. Different thresholds are set based on the statistical distribution of the moisture-to-greenness ratio index corresponding to monoculture and intercropping maize to identify intercropping maize samples and monoculture maize samples respectively. Separability analysis determined that the optimal time window for the Normalized Difference Vegetation Index (NDI) was 40 to 60 days before maize tasseling, and the optimal time window for shortwave infrared 1 band was 0 to 20 days after maize tasseling. A moisture-to-greenness ratio (MDR) greater than 2.0 was set as the threshold for identifying intercropped maize, corresponding to the 95th percentile of the WGR value distribution of monoculture maize; a MDR less than 1.0 was set as the threshold for identifying monoculture maize, corresponding to the 5th percentile of the WGR value distribution of intercropped maize.

[0024] The separability analysis uses the separability index to quantitatively assess the distinguishability of spectral or index features of two types of crops within a specific time window. The calculation formula is the ratio of the mean difference of the corresponding feature values ​​of the two types of crops to the sum of the standard deviations of the corresponding feature values ​​of the two types of crops by 1.96 times. The larger the value of the separability index, the better the separability. The time window with the largest value of the separability index is selected as the optimal time window.

[0025] S4. Combine the monoculture maize samples, intercropping maize samples and non-maize crop samples with a preset ratio obtained in step S3 into a training set, select multiple spectral bands and vegetation indices as input features, and train the classification model using the random forest algorithm. In the process of training the classification model using the random forest algorithm in S4, the parameters of the random forest algorithm model are set as follows: the number of decision trees is 200, and the number of features randomly selected when each decision tree grows is the square root of the total number of input features; the input features include the reflectance values ​​of the green band, red band, red-edge 1 band, near-infrared band, and short-wave infrared 1 band, as well as the normalized vegetation index and the converted chlorophyll absorption reflectance index.

[0026] S5. Input the extracted results of multiple spectral bands and vegetation indices of farmland areas in the study area into the trained classification model to obtain a preliminary classification result map; use the constructed validation dataset to calculate the classification accuracy index for validation; and combine regional agricultural statistics to perform area correction, and output the final spatial distribution map of monoculture and intercropping maize and the area estimation results.

[0027] The classification accuracy indicators include overall accuracy, producer accuracy, user accuracy, Kappa coefficient, and F1 score; the area correction adopts the linear scaling method to match and adjust the total national or regional area estimated by the model with the official statistical area of ​​the corresponding region; the classification model outputs a 10-meter resolution spatial distribution map of monoculture and intercropping maize.

[0028] like Figure 1As shown, this embodiment uses remote sensing mapping of monoculture and intercropping maize in Kenya as an example. Kenya is located in sub-Saharan Africa and is a major maize producer in Africa. Its terrain covers coastal plains, East African plateau, and the Great Rift Valley. The climate has a distinct gradient distribution and two planting seasons: a long rainy season (March-May) and a short rainy season (October-December). Small farms account for 75% of agricultural output, and monoculture and intercropping of maize coexist.

[0029] In this embodiment, 3,836 images from the long rainy season (March-August) of 2023 and 2,523 images from the short rainy season (November-February of the following year) of 2023-2024 were extracted from Sentinel-2 data, with an average of 27 and 14 observations per pixel, respectively. At the same time, Esri Land Cover, ESA World Cover and Google Dynamic World data were fused together, and erroneous plots were removed after field verification to form a reliable farmland mask.

[0030] In the stage of training the classification model using the random forest algorithm, 1,515 maize samples (891 monocultures and 299 intercroppings during the long rainy season; 222 monocultures and 103 intercroppings during the short rainy season) and 5,400 non-maize crop samples were obtained through field surveys. 70% of the non-maize samples were used for training, and 30% of the non-maize samples were used for validation.

[0031] Cloud removal, 10-day composite, and interpolation were performed on the images extracted from Sentinel-2 data to extract 5 spectral bands and 3 vegetation indices, forming a classification feature set. Specifically, the five spectral bands are green, red, red-edge 1, near-infrared, and shortwave infrared 1, and the three vegetation indices are optimized soil-adjusted vegetation index, converted chlorophyll absorption and reflectance index, and normalized vegetation index.

[0032] Next, the vegetation pigment variation index was calculated, and the optimal time window for vegetation and chlorophyll variation index was determined to be 10-30 days before heading, and the optimal time window for anthocyanin dynamic variation index was 40-60 days before heading. The corn pixels were then screened using a threshold of 0.05. Then, the moisture-to-greenness ratio index was calculated, and the optimal time window for the normalized vegetation index was determined to be 40-60 days before heading, and the optimal time window for the shortwave infrared spectral band 1 was 0-20 days after heading. Intercropping and monoculture maize samples were screened by setting a moisture-to-greenness ratio index threshold greater than 2.0 and a moisture-to-greenness ratio index threshold less than 1.0, respectively.

[0033] During the model training phase, the generated corn samples and 70% non-corn samples were used as the training set. The number of decision trees was set to 200, and the number of features was the square root of the total number of input features. Random forest models for long rainy season and short rainy season were trained separately. Specifically, based on 2438 long rainy season ground samples and 697 short rainy season ground samples, the overall accuracy, producer accuracy, user accuracy, Kappa coefficient, and F1 score were calculated to verify the model performance.

[0034] Finally, based on the acquired classification feature set, the trained classification model was used to classify monoculture maize and intercropped maize, generating a 10-meter resolution distribution map of monoculture and intercropped maize across Kenya. Based on ground sample validation, the overall accuracy was 88.74% during the long rainy season and 85.79% during the short rainy season, with F1 scores exceeding 75%. The area estimation results were adjusted in conjunction with agricultural statistics, showing good consistency with official statistics.

[0035] like Figure 2 The map shown shows the distribution of monoculture and intercropping maize in Kenya during the long rainy season in 2023. The generated map clearly presents the spatial distribution characteristics of monoculture and intercropping maize in Kenya: monoculture maize is mainly concentrated in highland areas with annual precipitation greater than 500 mm (accounting for 84%), while intercropping maize is distributed in a wider range of areas such as semi-arid regions, and the planting range during the short rainy season has expanded, reflecting the climate adaptation strategies of smallholder farmers.

[0036] This invention solves the core technical challenge of large-scale monoculture and intercropping maize mapping by organically combining phenologically assisted automated sample generation with machine learning classification. It features high precision, high automation, and strong adaptability, and can be extended to the mapping of complex planting systems in other small-scale agricultural regions around the world, providing technical support for food security and sustainable agricultural development.

[0037] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0038] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning, characterized in that, The method includes the following steps: S1. Acquire multispectral remote sensing images of the study area, perform cloud removal, temporal synthesis and missing value interpolation on the acquired multispectral remote sensing images, and extract multiple spectral bands and vegetation indices. S2. Multi-source land use data are fused to generate farmland masks, eliminating pixel interference from non-farmland areas in the study area, and extracting multiple spectral bands and vegetation indices of farmland areas in the study area. S3. Through field surveys, the center coordinates and growth stage information of monoculture maize, intercropped maize, and non-maize crops were recorded. Combined with high-resolution Google Earth imagery to supplement the samples, training and validation datasets were constructed. Separability analysis was used to determine the optimal time window for the vegetation pigment variation index, and thresholds were set based on the statistical distribution of the vegetation pigment variation index to filter maize pixels. Separability analysis was also used to determine the optimal time window for the moisture and greenness ratio index, distinguishing between monoculture and intercropped maize pixels. Different thresholds were set based on the statistical distribution of the moisture and greenness ratio indices corresponding to monoculture and intercropped maize to identify intercropped and monoculture maize samples respectively. S4. Combine the monoculture maize samples, intercropping maize samples and non-maize crop samples with a preset ratio obtained in step S3 into a training set, select multiple spectral bands and vegetation indices as input features, and train the classification model using the random forest algorithm. S5. Input the extracted results of multiple spectral bands and vegetation indices of farmland areas in the study area into the trained classification model to obtain a preliminary classification result map; use the constructed validation dataset to calculate the classification accuracy index for validation; and combine regional agricultural statistics to perform area correction, and output the final spatial distribution map of monoculture and intercropping maize and the area estimation results.

2. The remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning according to claim 1, characterized in that: The multiple spectral bands extracted in S1 include the reflectance values ​​of the green band, red band, red-edge band 1, short-wave infrared band 1, and near-infrared band. The vegetation index includes the soil-optimized vegetation index, the converted chlorophyll absorption and reflectance index, and the normalized vegetation index. The formula for calculating the optimized soil-adjusted vegetation index is: OSAVI = 1.16 × (NIR - R) / (NIR + R + 0.16), where OSAVI represents the optimized soil-adjusted vegetation index; NIR represents the reflectance value of the near-infrared band at the corresponding time point; and R represents the reflectance value of the red band at the corresponding time point. The formula for calculating the converted chlorophyll absorption reflectance index is: TCARI=3×[(RE1-R)-0.2×(RE1-G)×RE1 / R], where TCARI represents the converted chlorophyll absorption reflectance index; RE1 represents the reflectance value of the red-edge band 1 at the corresponding time point; and G represents the reflectance value of the green band at the corresponding time point. The formula for calculating the Normalized Difference Vegetation Index (NDVI) is: NDVI = (NIR - R) / (NIR + R), where NDVI represents the Normalized Difference Vegetation Index.

3. The remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning according to claim 1, characterized in that: The farmland mask is generated by fusing Esri land cover data, ESA world cover data and Google dynamic world data, and the fusion results are subjected to intersection or union operations, and then visually verified by high-resolution images and manually corrected to remove non-farmland plots.

4. The remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning according to claim 1, characterized in that: The vegetation pigment variation index is obtained by multiplying the vegetation and chlorophyll variation index by the anthocyanin dynamic variation index. The optimal time window corresponding to the vegetation and chlorophyll variation index and the anthocyanin dynamic variation index is determined by the separability analysis. The threshold is set based on the statistical distribution of the vegetation pigment variation index of non-corn crops to screen out corn pixels. The vegetation and chlorophyll variation index is calculated by accumulating the absolute values ​​of the differences between the optimized soil-adjusted vegetation index and the converted chlorophyll absorption and reflection index at each time point after normalization during the key phenological period of maize. The optimized soil-adjusted vegetation index and the converted chlorophyll absorption and reflection index are normalized in the same way, and the normalized values ​​of the optimized soil-adjusted vegetation index and the converted chlorophyll absorption and reflection index are both in the range of [0, 1]. The anthocyanin dynamic variation index is calculated by accumulating the absolute values ​​of the first difference of the anthocyanin reflectance index at each time point during the key phenological period of maize.

5. The remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning according to claim 4, characterized in that: The optimal time window for vegetation pigment variation index was determined to be 10 to 30 days before corn tasseling, and the optimal time window for anthocyanin dynamic variation index was determined to be 40 to 60 days before corn tasseling, based on separability analysis. The threshold for the vegetation pigment variation index was set to 0.05, which is the 95th percentile of the distribution of vegetation pigment variation index values ​​for non-corn crops.

6. The remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning according to claim 1, characterized in that: The moisture-to-greenness ratio index is the ratio of the mean of the shortwave infrared 1 band to the mean of the normalized vegetation index. The optimal time window for the normalized vegetation index and the shortwave infrared 1 band is determined by the separability analysis. Different thresholds are set based on the statistical distribution of the moisture-to-greenness ratio index corresponding to monoculture and intercropping maize to identify intercropping maize samples and monoculture maize samples respectively. Separability analysis determined that the optimal time window for the Normalized Difference Vegetation Index (NDI) was 40 to 60 days before maize tasseling, and the optimal time window for shortwave infrared 1 band was 0 to 20 days after maize tasseling. A moisture-to-greenness ratio (MDR) greater than 2.0 was set as the threshold for identifying intercropped maize, corresponding to the 95th percentile of the WGR value distribution of monoculture maize; a MDR less than 1.0 was set as the threshold for identifying monoculture maize, corresponding to the 5th percentile of the WGR value distribution of intercropped maize.

7. The remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning according to claim 1, characterized in that: The separability analysis uses the separability index to quantitatively assess the distinguishability of spectral or index features of two types of crops within a specific time window. The calculation formula is the ratio of the mean difference of the corresponding feature values ​​of the two types of crops to the sum of the corresponding standard deviations of the feature values ​​of the two types of crops, which is 1.96 times. The time window with the largest separability index value is selected as the optimal time window.

8. The remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning according to claim 1, characterized in that: In the process of training the classification model using the random forest algorithm in S4, the parameters of the random forest algorithm model are set as follows: the number of decision trees is 200, and the number of features randomly selected when each decision tree grows is the square root of the total number of input features; the input features include the reflectance values ​​of the green band, red band, red-edge 1 band, near-infrared band, and short-wave infrared 1 band, as well as the normalized vegetation index and the converted chlorophyll absorption reflectance index.

9. The remote sensing mapping method for monoculture and intercropping maize based on phenological assistance and machine learning according to claim 1, characterized in that: The classification accuracy indicators include overall accuracy, producer accuracy, user accuracy, Kappa coefficient, and F1 score; the area correction adopts the linear scaling method to match and adjust the total national or regional area estimated by the model with the official statistical area of ​​the corresponding region.

Images