Crop classification method based on multi-angle satellite observation data

Abstract

The present invention relates to the technical field of crop classification, and in particular to a crop classification method based on multi-angle satellite observation data, comprising S1. acquiring zenith reflection data and multi-angle observation factor data of a study area; S2. using a land cover data set to screen crops in the zenith reflection data and multi-angle observation factor data, and obtaining single-angle observation type crop images and multi-angle observation type crop images; S3. constructing a sampling point database of different types of crops in the study area; S4. constructing a training data set and a verification data set based on the sampling point database, single-angle observation type crop images and multi-angle observation type crop images, and training four machine learning classifiers provided by a GEE platform; S5. using the trained four machine learning classifiers to classify single-angle observation type crop images and multi-angle observation type crop images. The present invention can improve the accuracy of crop distribution mapping.

Description

Technical Field

[0001] The present invention relates to the technical field of crop classification, and in particular to a crop classification method based on multi-angle satellite observation data. Background Art

[0002] Accurate crop classification and distribution mapping are essential for crop growth monitoring and yield prediction. Remote sensing monitoring is a reliable method for agricultural production. Due to its advantages of high frequency, large scale and spatial consistency, it can effectively monitor agricultural production to obtain various details about crop production from local to large areas, such as phenological status, crop health, crop type and crop yield assessment.

[0003] The cumulative number of remote sensing satellites generates up to 1PB of unprocessed images each year. This increase in the amount and scope of data, known as the big data dilemma, introduces new tasks in data management and requires new methods to extract specific data from remote sensing data science. Google Earth Engine (GEE) has become a key platform for large-scale land cover classification and crop type mapping, with its powerful computing power, rich satellite imagery, and various machine learning classifiers.

[0004] Machine learning methods are pioneering methods for processing remote sensing data. Machine learning methods can discover unique indicators of crops and are widely used in crop classification, crop mapping, and yield estimation. For example: The GEE platform provides several non-parametric machine learning classifiers for classification. Although these machine learning classifiers have been well applied in single-angle satellite images. However, since single-angle satellite images cannot fully capture the structural characteristics of different crops, the application of single-angle satellite images in crop classification has limitations, and images from a single angle can only obtain information under specific observation conditions, which may cause the image to contain interference factors related to the observation angle, thereby increasing the risk of classification errors. Summary of the invention

[0005] In view of this, the present invention aims to provide a crop classification method based on multi-angle satellite observation data to solve the problems existing in the existing machine learning classifier using single observation angle satellite images.

[0006] A crop classification method based on multi-angle satellite observation data comprises the following steps:

[0007] S1. Obtain remote sensing data from the MODIS sensor from the GEE platform and limit it to the study area through a spatial filter; the remote sensing data includes zenith reflectance data and multi-angle observation factor data;

[0008] S2. The land cover dataset of the MODIS sensor is used as a mask for crop pixels, and crops in the zenith reflectance data and multi-angle observation factor data are screened according to the classification of the International Geosphere-Biosphere Program to obtain single-angle observation type crop images and multi-angle observation type crop images;

[0009] S3. Build a database of sampling points for different types of crops in the study area;

[0010] S4. The sampling point database, single-angle observation type crop images and multi-angle observation type crop images are used as training samples and verification samples to construct training data sets and verification data sets respectively, and the four machine learning classifiers provided by the GEE platform are trained; the four machine learning classifiers are random forest, classification and regression tree, support vector machine and naive Bayes;

[0011] S5. Use the four trained machine learning classifiers to perform crop classification on single-angle observation type crop images and multi-angle observation type crop images.

[0012] Furthermore, zenith reflectance data are provided by the MCD43A4 V6 product, and multi-angle observation factor data are provided by the MCD43A1V6 product.

[0013] Furthermore, a date filter is used to limit the zenith reflectance data and multi-angle observation factor data to a certain year, and the growing period and harvest period of the crops are selected. The median of the zenith reflectance data and multi-angle observation factor data in the two periods are provided as time measurements to four machine learning classifiers.

[0014] Furthermore, the land cover dataset is provided by the MCD12Q1 V6 product.

[0015] Furthermore, spatial connection is performed by removing training samples within 1000 meters of each validation sample.

[0016] Furthermore, the sampling points of each machine learning classifier input are the same.

[0017] Furthermore, a sampling point database was constructed using the field sampling point verification method.

[0018] Furthermore, after step S5, the method further includes the following steps:

[0019] S6. Construct a crop classification database based on the crop classification results and draw a crop distribution map.

[0020] Further, crops include corn, rice and soybeans.

[0021] Compared with the prior art, the present invention can achieve the following technical effects:

[0022] 1. This method uses the multi-angle observation data of the MODIS sensor to classify crops. The multi-angle observation data can capture the structural characteristics of crops. Specifically, the bidirectional reflectance distribution function and albedo model parameter data provided by the MCD43A1 V6 product are linearly fitted through the kernel driving function. Coefficients related to the crop structural characteristics but independent of the observation angle are extracted from the reflectivity data observed from multiple angles. These coefficients are used for classification, which can reduce the classification error caused by changes in the observation angle of the MODIS sensor.

[0023] 2. Compared with the crop distribution map produced based on single-angle observation data, the crop distribution map produced by this method based on multi-angle observation data is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A flowchart of a crop classification method based on multi-angle satellite observation data according to an embodiment of the present invention;

[0025] Figure 2 A technical roadmap for the crop classification method based on multi-angle satellite observation data described in the embodiment of the present invention;

[0026] Figure 3 A MODIS composite image of the study area described in the embodiment of the present invention;

[0027] Figure 4-Figure 11 The crop distribution maps are respectively drawn according to the results of crop classification of zenith reflection data and multi-angle observation factor data based on four machine learning classifiers as described in the embodiments of the present invention. DETAILED DESCRIPTION

[0028] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the following description, the same modules are represented by the same reference numerals. In the case of the same reference numerals, their names and functions are also the same. Therefore, the detailed description thereof will not be repeated.

[0029] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.

[0030] The present invention provides a method for crop classification using multi-angle bidirectional reflectance distribution function observation information. The method uses MODIS multi-angle observation data and adopts multiple machine learning classifiers to classify crops, and finally constructs a crop classification database, thereby realizing high-precision crop distribution mapping.

[0031] The present invention will be described in detail below with reference to the accompanying drawings and taking Jilin Province as a research area.

[0032] like Figure 1 and Fig.11 As shown, the crop classification method based on multi-angle satellite observation data provided by the present invention includes the following steps:

[0033] S1. Obtain remote sensing data from the MODIS sensor from the GEE platform and limit it to the study area through a spatial filter; the remote sensing data includes zenith reflectance data and multi-angle observation factor data.

[0034] The MODIS sensor, or Moderate-resolution Imaging Spectroradiometer (MODIS), is a large-scale space remote sensing instrument developed by NASA to understand global climate changes and the impact of human activities on the climate.

[0035] The present invention uses two MODIS reflectivity products, namely the reflectivity data adjusted by the downward-viewing bidirectional reflectance distribution function (i.e., MCD43A4 V6 product) and the bidirectional reflectance distribution function and albedo model parameter data (i.e., MCD43A1V6 product). The MCD43A4 V6 product and the MCD43A1 V6 product are part of the MODIS bidirectional reflectance distribution function / albedo suite and provide observations of the anisotropic reflectance properties of the earth's surface, which is essential for various applications such as vegetation monitoring, land cover mapping, and climate modeling. The wavelength and spatial resolution of the MODIS sensor are shown in Table 1. The MCD43A4 V6 product is fitted using the bidirectional reflectance distribution function to simulate the values ​​observed as if in the zenith observation direction. The MCD43A1 V6 product provides three model weight parameters for each MODIS band (isotropic, volumetric and geometric parameters). The bidirectional reflectance distribution function and albedo model parameter data provided by the MCD43A1 V6 product are linearly fitted through the kernel driving function. Coefficients related to crop structural characteristics but not related to the observation angle are extracted from the reflectance data observed from multiple angles. These coefficients are used for classification, which can reduce the classification error caused by changes in the observation angle of the MODIS sensor. The MCD43A4 V6 product and the MCD43A1 V6 product are generated by combining satellite data and modeling technology.

[0036] Table 1 Channels (wavelengths) of MODIS sensors

[0037]

[0038] The present invention classifies crops by using remote sensing data observed at single angle and multiple angles. Single-angle observation remote sensing data is the reflectivity data adjusted by the downward-viewing bidirectional reflectance distribution function of the MODIS sensor (provided by the MCD43A4V6 product), which is called zenith reflectance data. Multi-angle observation remote sensing data is the bidirectional reflectance distribution function and albedo model parameter data of the MODIS sensor (provided by the MCD43A1 V6 product), which is called multi-angle observation factor data.

[0039] After obtaining the zenith reflectance data and multi-angle observation factor data from the GEE platform, a spatial filter is used to limit the data to Jilin Province, and a date filter is used to limit the data to 2017. All available zenith reflectance data and multi-angle observation factor data of two different periods of crops are provided to a machine learning classifier (hereinafter referred to as the classifier). Specifically, the growing period (170th day to 230th day of 2017) and the harvest period (231st day to 291st day of 2017) of the crops are selected. During the growing period and the harvest period, the median of all band data (i.e., zenith reflectance data and multi-angle observation factor data) is provided to the classifier as a time metric. The time metric improves the classification accuracy by 7%. Due to the asymmetric error distribution, the present invention uses the median instead of the mean.

[0040] For zenith reflection data, the present invention has 7 (band) × 1 (direction) × 2 (stage) × 1 (time measurement) = 14 classification indexes provided to the classifier for supervised classification. For multi-angle observation factor data, the present invention has 7 (band) × 3 (direction) × 2 (stage) × 1 (time measurement) = 42 classification indexes provided to the classifier for supervised classification.

[0041] S2. The land cover dataset of the MODIS sensor is used as a mask for crop pixels, and crops in the zenith reflectance data and multi-angle observation factor data are screened according to the classification of the International Geosphere-Biosphere Program to obtain single-angle observation type crop images and multi-angle observation type crop images.

[0042] The land cover dataset is provided by the MCD12Q1 V6 product. The MCD12Q1 V6 product is a global annual land cover type product from the MODIS sensor. This product provides global land cover classification data with a spatial resolution of 500m. The land cover dataset provided by the MCD12Q1 V6 product is generated once a year and contains information on various land cover types, such as urban areas, grasslands, farmlands, wetlands, and forests. The MCD12Q1 V6 product uses a variety of classification schemes that help scientists and decision makers better understand land cover dynamics, monitor changes, and assess the impacts of land use and land cover changes on ecosystems, climate, and human-environment interactions. The classification of the International Geosphere-Biosphere Program (land cover type 1) is used as a mask to select farmland pixels in Jilin Province. The area of ​​the pixels selected by the mask is not less than 60% of cultivated land.

[0043] Crop pixels are filtered out by applying the 2017 MCD12Q1 V6 product as a mask.

[0044] S3. Build a database of sampling points for different types of crops in the study area.

[0045] In 2017, three field surveys were conducted in Jilin Province to obtain rich and reliable prior knowledge and validation information about the distribution of crops within Jilin Province. The sampling points were selected using a field sampling point validation method, which involved a gradual journey from the northwest to the southeast of Jilin Province. The field experiment was divided into three main phases: the first phase (September 1-6) focused on the northwest of Jilin Province, mainly in Baicheng City and its surrounding areas; the second phase (September 12-21) focused on the central region of Jilin Province, mainly in Changchun City and its surrounding areas; the third phase (October 1-5) mainly examined the southeastern region of Jilin Province, mainly in Jilin City and its surrounding areas. After completing each field experiment phase, return to the laboratory to process and validate the collected field data. Once the data are validated, the subsequent field work phase will be initiated.

[0046] Before starting the field experiment, sampling points were selected indoors roughly using high spatial resolution imagery (i.e., Sentinel-2) and predicted crop types, giving priority to areas that were as large as possible, minimally disturbed, and easily accessible, while ensuring that all sampling points were evenly distributed within the study area. Once all sampling points were identified, a portable global positioning system (GPS) was used to measure the geometric positions, verify the intended crop types, and record the main information for each field. GPS measurements were made as close to the center of the field as possible, and preliminary data were recorded at the same time. The main data mainly included the determined crop type, altitude, latitude, and longitude. A total of 258 sampling points for corn, 132 for rice, and 53 for soybeans were obtained. The distribution map of the field sampling points is shown in the figure. Figure 3As shown, the base map is a MODIS composite image during the growing period (day 170-day 230 of the year). The RGB color consists of bands 1 (red), 4 (green), and 3 (blue) with median reflectance. The triangle, square, and circle symbols represent the sampling points for corn, rice, and soybean, respectively.

[0047] S4. The sampling point database, single-angle observation type crop images and multi-angle observation type crop images are used as training samples and verification samples to construct training data sets and verification data sets respectively, and the four classifiers provided by the GEE platform are trained; among them, the four classifiers are random forest, classification and regression tree, support vector machine and naive Bayes.

[0048] The ratio of training data set to validation data set is 7:3, and spatial connection is performed by removing training samples within 1000 meters of each validation sample to avoid correlation due to spatial autocorrelation of the predicted phenomenon. The sampling points of each classifier input are the same.

[0049] S5. Use the four trained classifiers to classify crops for single-angle observation type crop images and multi-angle observation type crop images.

[0050] After crop classification, the classification accuracy was verified based on the input data and classifier type.

[0051] The training input parameters of the classifier are shown in Table 2:

[0052] Table 2 Training input parameters for each classifier

[0053]

[0054] The training input parameters in Table 2 were determined by considering the GEE official help document.

[0055] After step S5, the following steps are also included:

[0056] S6. Construct a crop classification database based on the crop classification results and draw a crop distribution map.

[0057] The present invention compares zenith reflection data and multi-angle observation factor data, and analyzes the classification results and precisions of four classifiers.

[0058] Crop distribution map Figure 4-Figure 11 As shown, Figure 4 and Figure 5 The crop distribution maps are drawn by classifying the zenith reflectance data and multi-angle observation factor data using random forest classifiers; Figure 6 and Figure 7The crop distribution maps are drawn by classifying the zenith reflectance data and multi-angle observation factor data using classification and regression tree classifiers respectively; Figure 8 and Fig. 9 The crop distribution maps are drawn by classifying the zenith reflectance data and multi-angle observation factor data using the support vector machine classifier; Fig.10 and Fig.11 The crop distribution maps are drawn by classifying the zenith reflectance data and multi-angle observation factor data using the naive Bayes classifier.

[0059] from Figure 4-Figure 11 As can be seen in Table 3, the location of corn shows good consistency in the classification results of the four classifiers, while the location of soybean varies greatly. Table 3 shows the crop area and the corresponding classification accuracy. The choice of classifier has a greater impact on the results than the input data. Different classifiers produced significantly different classification results for the same classification data, with the maximum standard deviation of corn area being 16556.75 square kilometers. In contrast, different classification data using the same classifier produced less different results, with the maximum standard deviation of corn area being 4083.54 square kilometers. It is worth noting that the support vector machine classifier cannot classify soybean pixels using zenith reflectance data, but can classify soybean pixels using multi-angle observation factor data (although the occupied area is very small). The naive Bayes classifier cannot classify soybean and wheat pixels, whether using zenith reflectance data or multi-angle observation factor data.

[0060] Table 3 Crop regions and classification accuracy

[0061]

[0062] The present invention proposes to use additional information from input parameters, such as bidirectional optical properties of the ground, to compensate for the uncertainty caused by spectral confusion problems in land cover classification and improve accuracy. For example, the land cover classification product (MCD12Q1 V6) of the Moderate Resolution Imaging Spectroradiometer (MODIS) uses multi-angle satellite observation data for land cover classification. Considering the diversity of crop structures, multi-angle optical observation satellite images help improve the accuracy of crop mapping.

[0063] Therefore, the present invention proposes to use the bidirectional reflectance distribution function features captured by multi-angle observations to improve the accuracy of crop mapping. To verify this conclusion, the present invention evaluates the accuracy of four classifiers by comparing their performance in crop mapping using zenith reflectance data and multi-angle observation factor data.

[0064] The accuracy assessment was performed from three different perspectives, namely, the classification accuracy assessment between classifiers, the assessment of multi-angle observation data, and the accuracy assessment between crop types. The evaluation criteria were overall accuracy and the Kappa coefficient. Overall accuracy is a metric that measures the ratio of correctly classified pixels to the total number of pixels in the image. This is a simple and easy-to-understand metric that provides an overall estimate of the robustness of the classification model. On the other hand, the Kappa coefficient is a statistical metric that takes into account the agreement between the observed and expected classifications and corrects for the agreement expected by chance. The Kappa coefficient ranges from -1 to 1, where 1 indicates perfect agreement, 0 indicates probability agreement, and -1 indicates complete disagreement. Because the agreement expected by chance is taken into account, the Kappa coefficient is more reliable than the overall accuracy. Therefore, the Kappa coefficient is a more reliable measure of the performance of the classification model.

[0065] Table 4 Producer accuracy and user accuracy of training dataset

[0066]

[0067] Table 5 Producer accuracy and user accuracy of the validation dataset

[0068]

[0069] (1) Classification accuracy evaluation among classifiers: Table 3 shows the classification results and accuracy (overall accuracy and Kappa coefficient) of the random forest classifier, classification and regression tree classifier, support vector machine classifier, and naive Bayes classifier on the training and validation data. The four classifiers showed noteworthy differences in performance. The classification and regression tree classifier had the highest training accuracy (overall training accuracy equal to 98.7%, training Kappa coefficient equal to 0.977). The random forest classifier had the highest overall validation accuracy (overall validation accuracy exceeded 75%, Kappa coefficient exceeded 0.500). The training and validation accuracy of the support vector machine classifier (overall accuracy exceeded 73.0%, Kappa coefficient exceeded 0.440) was lower than that of the classification and regression tree and random forest. The naive Bayes classifier had the worst accuracy among the four classifiers (overall validation accuracy equal to 54.9%, Kappa coefficient equal to 0.000).

[0070] The choice of classifier can greatly affect the accuracy of classification. In the present invention, when using MODIS remote sensing images, the classifier is the main factor affecting the classification results, and the zenith reflectance data or multi-angle observation factor data have less impact on the results. Among the classifiers provided by the GEE platform, the random forest classifier has the highest validation accuracy, while the classification and regression tree classifier has the highest training accuracy. The support vector machine classifier and the random forest classifier perform similarly and well in validation accuracy. The naive Bayes classifier performs the worst. The random forest classifier is generally considered to be the most accurate classifier of the four classifiers provided by the GEE platform. For example, a meta-analysis of 349 publications on the GEE platform in the past decade showed that the random forest algorithm is the most commonly used classifier for satellite image classification. In the optical data case of the present invention, random forests showed the highest validation accuracy. However, the training accuracy of the classification and regression tree classifier is higher than that of the random forest classifier. This may be explained by several studies on crop mapping and tree species mapping. Classification and regression tree classifiers are suitable for areas with simple land cover classes and consistent spatial distribution, while random forest classifiers are suitable for areas with complex land cover types and scattered spatial distribution. In addition, the quality of training samples can also affect the performance of classifiers. Compared with previous studies, the spatial resolution of MODIS of 500 meters may not be sufficient for crop mapping studies. Therefore, the components within a pixel may be more complex and the mixed pixel effect may be more serious, affecting the quality of training samples. Noise in the training data also reduces the accuracy of the support vector machine. Compared with the random forest classifier, the support vector machine classifier may be more sensitive to the quality of training data. Therefore, consistent with the crop mapping literature using multi-temporal Sentinel-2 images, the accuracy of the support vector machine classifier is lower than that of the random forest classifier. The naive Bayes classifier performs the worst because the naive Bayes classifier can be easily deceived by irrelevant features. Therefore, the present invention helps to select the importance of classifiers based on the intrinsic properties of the research target and the type of image data.

[0071] (2) Evaluation of multi-angle observation data: The use of multi-angle observation data can improve the classification accuracy of the classifier. Specifically, the accuracy of the random forest classifier, classification and regression tree classifier, and support vector machine classifier using multi-angle observation factor data is more improved than that using zenith reflectance data. The overall validation accuracy of the random forest classifier increased from 75.5% to 79.4%, and the validation kappa coefficient increased from 0.539 to 0.602. The overall validation accuracy of the classification and regression tree classifier increased from 58.8% to 63.7%, and the validation kappa coefficient increased from 0.307 to 0.368. The overall validation accuracy of the support vector machine classifier increased from 73.5% to 77.5%, and the validation kappa coefficient increased from 0.450 to 0.542. However, the accuracy difference caused by using zenith reflectance data and multi-angle observation factor data (overall accuracy up to 4.9%, kappa coefficient up to 0.092) is smaller than the difference caused by using different classifiers (overall accuracy up to 43.0%, kappa coefficient up to 0.977). Therefore, the choice of classifier is the main factor affecting the classification accuracy (as shown in Table 3).

[0072] Multi-angle optical observations have been widely used in earth observations because they can capture bidirectional reflectance distribution function features, and it has been shown that bidirectional reflectance distribution function features can help with land cover classification and vegetation structure (e.g., leaf area index, leaf inclination distribution, and leaf size) inversion. The spectral variability between crops is much lower than the spectral variability between land covers, which increases the difficulty of classification. However, the structure of crops (e.g., leaf area index, leaf inclination distribution, and leaf size) may vary greatly, which can be well expressed by bidirectional reflectance distribution function features. Therefore, the present invention proposes to use bidirectional reflectance distribution function features captured by multi-angle optical observation data for crop classification. The results of the present invention show that in most cases, the use of multi-angle optical observations can improve classification accuracy compared to using only zenith reflectance data.

[0073] (3) Accuracy evaluation between crop types: Tables 4 and 5 show the producer accuracy and user accuracy metrics for the training dataset and validation dataset, respectively. Producer accuracy represents the number of pixels correctly classified out of the total number of pixels of a specific category in the reference data. Producer accuracy evaluates the ability of the classifier to accurately identify a specific category, indicating the probability that a certain category in the reference data (ground truth) is correctly classified in the image classification. This metric is also known as the commission error. A high producer accuracy indicates that the classifier effectively identifies the specific category, while a low producer accuracy indicates that the classifier may misclassify the category. User accuracy represents the ratio of the number of pixels correctly classified for a given category to the total number of pixels classified as that category in the image classification. This metric is also known as the omission error. User accuracy evaluates the reliability of the classification of a specific category, indicating the probability that a certain category in the classified image corresponds to the same category in the reference data (ground truth). A high user accuracy indicates that the classification result of a specific category is reliable, while a low user accuracy indicates that the classifier may include pixels of other categories in the specific category. The producer accuracy and user accuracy values ​​for the training dataset were higher than those for the validation dataset, with average values ​​of 68.22% and 46.74%, respectively. Among the three crop types, corn had the highest accuracy (83.42%), followed by rice (62.05%), while soybean had the lowest accuracy (26.97%). It is worth noting that only the classification and regression tree classifiers produced reliable classification results for soybean.

[0074] It has been reported that increasing the number of training data sets can moderately improve the performance of the classifier. In the classification results of the present invention, corn has the highest classification accuracy, followed by rice, while soybean has the lowest accuracy. This order corresponds to the number of sampling points of the present invention. However, when the training sample size approaches a certain threshold, the effect of increasing the training sample size on improving the classification accuracy will gradually decrease, and even reach a bottleneck. At this time, other methods should be used to improve the classification accuracy, such as improving feature selection, adjusting classifier parameters, etc. Since the main crops dominate in number in the training samples, it is a challenge to classify crops that do not dominate in number.

[0075] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the disclosure of the present invention can be performed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document does not limit this.

[0076] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A crop classification method based on multi-angle satellite observation data, characterized in that: The steps include: S1. Obtain remote sensing data from the MODIS sensor from the GEE platform and limit it to the study area through a spatial filter; the remote sensing data includes zenith reflectance data and multi-angle observation factor data; S2. The land cover dataset of the MODIS sensor is used as a mask for crop pixels, and crops in the zenith reflectance data and multi-angle observation factor data are screened according to the classification of the International Geosphere-Biosphere Program to obtain single-angle observation type crop images and multi-angle observation type crop images; S3. Build a database of sampling points for different types of crops in the study area; S4. The sampling point database, single-angle observation type crop images and multi-angle observation type crop images are used as training samples and verification samples to construct training data sets and verification data sets respectively, and the four machine learning classifiers provided by the GEE platform are trained; the four machine learning classifiers are random forest, classification and regression tree, support vector machine and naive Bayes; S5. Use the four trained machine learning classifiers to classify crops for single-angle observation type crop images and multi-angle observation type crop images.

2. The crop classification method based on multi-angle satellite observation data according to claim 1, characterized in that: Zenith reflectance data are provided by MCD43A4 V6 product, and multi-angle observation factor data are provided by MCD43A1 V6 product.

3. The crop classification method based on multi-angle satellite observation data according to claim 2 is characterized in that: A date filter is used to limit the zenith reflectance data and multi-angle observation factor data to a certain year. The growing period and harvest period of the crops are selected, and the median of the zenith reflectance data and multi-angle observation factor data in the two periods is provided as a time measure to four machine learning classifiers.

4. The crop classification method based on multi-angle satellite observation data according to claim 1, characterized in that: The land cover dataset is provided by the MCD12Q1 V6 product.

5. The crop classification method based on multi-angle satellite observation data according to claim 1, characterized in that: Spatial concatenation was performed by removing training samples that were within 1000 m of each validation sample.

6. The crop classification method based on multi-angle satellite observation data according to claim 1, characterized in that: The sampling points of each machine learning classifier input are the same.

7. The crop classification method based on multi-angle satellite observation data according to claim 1, characterized in that: The sampling point database was constructed using the field sampling point verification method.

8. The crop classification method based on multi-angle satellite observation data according to claim 1, characterized in that: After step S5, the following steps are also included: S6. Construct a crop classification database based on the crop classification results and draw a crop distribution map.

9. The crop classification method based on multi-angle satellite observation data according to any one of claims 1 to 8, characterized in that: Crops include corn, rice and soybeans.

Images