A timing radar crop classification method based on a consistency coefficient

Abstract

The application discloses a kind of consistency coefficient-based timing radar crop classification method, belong to remote sensing crop identification and classification technical field, the method includes: obtaining the timing polarized synthetic aperture radar image covering target area, and pre-processing obtains the timing backscattering coefficient data under different polarization modes;According to training sample, standard timing reference curve under each polarization mode of each land object category is constructed;The timing backscattering curve of the pixel to be classified is respectively consistent with the standard timing reference curve corresponding to each land object category Correlation analysis is carried out, and the consistency correlation coefficient corresponding to each category is obtained;According to the consistency correlation coefficient under different polarization modes, classification feature vector is constructed, and the crop classification model is trained using classification feature vector;The target area is classified and predicted using the crop classification model trained, and the crop classification result is obtained.The application uses consistency correlation feature to represent the timing change characteristics of land object, which provides a new feature construction method for crop classification.

Description

Technical Field

[0001] This application relates to the field of remote sensing crop identification and classification technology, and in particular to a time-series radar crop classification method based on consistency coefficient. Background Technology

[0002] Information on crop planting area and spatial distribution is crucial foundational data for ensuring national food security, formulating agricultural policies, and regulating agricultural production structure. Traditional crop area statistics rely on manual surveys or ground quadrat surveys, which are inefficient and struggle to obtain detailed spatial distribution data over large areas. With the rapid development of remote sensing Earth observation technology, satellite remote sensing has gradually become the mainstream technology for crop identification and area extraction. Compared to optical remote sensing, which is susceptible to cloud, rain, and fog and struggles to acquire effective images during critical crop growth periods, Synthetic Aperture Radar (SAR) offers advantages such as all-weather, all-time availability, strong penetration, and short revisit cycles. In particular, high spatial resolution polarimetric SAR data provides an opportunity for large-scale dynamic monitoring and classification of crops. Currently, the mainstream methods for crop classification based on SAR time-series imagery can be summarized into two main categories. The first category is decision tree or rule set methods based on phenological characteristics. These methods require a deep understanding of the phenological calendars of different crops in specific regions (such as sowing, tillering, heading, and maturity stages). Based on the rising and falling patterns of backscattering coefficients during each phenological stage, classification rules are established for different crops. However, these methods heavily rely on prior phenological knowledge; phenological differences across regions and years can cause these rules to fail, resulting in extremely low portability and automation. The second category is machine learning methods based on SAR temporal features. These methods extract temporal backscattering coefficients, polarization decomposition parameters, and texture features from SAR images, inputting them into classifiers such as support vector machines and random forests for supervised classification to identify crops. However, most existing methods directly use the original backscattering coefficients or their simple transformations (such as polarization ratios and differences) as features, and the types of feature parameters are numerous. Because SAR backscattering is affected by various factors such as soil dielectric constant, surface roughness, crop water content and biomass, the original time-series scattering curves of different land cover categories often have severe nonlinear aliasing, resulting in insufficient separability between categories. This is especially true for different varieties of rice (such as early rice and mid-season rice) or high-water-content vegetation (such as ramie and forage) with extremely similar scattering characteristics, making it difficult to meet the needs of practical operational applications in terms of classification accuracy.

[0003] Furthermore, existing methods rarely fully utilize the temporal stability information of SAR imagery when constructing features. The Concordance Correlation Coefficient (CCC), as an indicator for evaluating the consistency between two sets of data, has the dual advantages of measuring correlation and accuracy, and has been applied in meteorology and hydrology in recent years. However, its application in temporal SAR crop classification, constructing CCC feature maps based on standard reference curves, and integrating multi-polarized CCC features for ensemble learning classification is currently a blank. Summary of the Invention

[0004] This invention addresses the technical problems existing in the background art by proposing a time-series radar crop classification method based on consistency coefficients. This method can fundamentally eliminate the dependence on prior phenological knowledge of crops and reduce the types of required feature parameters. By mining the inherent consistency characteristics of ground objects in time, it can achieve high-precision and highly robust automatic crop classification.

[0005] To solve the technical problem, the technical solution of the present invention is as follows:

[0006] A time-series radar crop classification method based on consistency coefficient includes the following steps:

[0007] S1. Acquire temporal polarimetric synthetic aperture radar (TPSAR) images covering the target area, and preprocess the temporal polarimetric synthetic aperture radar (TPSAR) images to obtain temporal backscattering coefficient data under different polarization modes.

[0008] S2. Construct a training sample set based on the time-series backscattering coefficient data, and construct a standard time-series reference curve for each land cover category under each polarization mode based on the time-series backscattering curves corresponding to each land cover category in the training sample set.

[0009] S3. For the pixel to be classified, perform consistency correlation analysis between its time-series backscattering curve under each polarization mode and the standard time-series reference curve corresponding to each land cover category to obtain the consistency correlation coefficient of the pixel to be classified corresponding to each land cover category.

[0010] S4. Construct the classification feature vector of the pixel to be classified based on the consistency correlation coefficient obtained under different polarization modes;

[0011] S5. Train a crop classification model using the classification feature vectors corresponding to the training sample set;

[0012] S6. Using the trained crop classification model, predict the classification feature vectors corresponding to the pixels to be classified in the target area to obtain the crop classification results for the target area.

[0013] Furthermore, in step S1, the preprocessing includes orbit correction, radiometric correction, multi-view processing, filtering, geocoding, and backscattering coefficient linearization to obtain time-series linear backscattering coefficient data under different polarization modes.

[0014] Furthermore, step S2 includes:

[0015] Based on the preset land cover categories, the temporal backscattering curves of training pixels of each category under each polarization mode are extracted from the training sample set;

[0016] For each land cover category, under each polarization mode, the backscattering of training pixels of the same category is averaged in each time phase. That is, the temporal backscattering curve of training pixels for each land cover category is averaged phase by phase, and the average values ​​of each phase are used to construct a standard time-series reference curve. The standard time-series reference curve corresponding to this land cover category is obtained.

[0017] Furthermore, in step S3, the consistency correlation analysis is performed using the consistency correlation coefficient, which comprehensively reflects the degree of correlation and consistency between the temporal backscattering curve of the pixel to be classified and the standard temporal reference curve.

[0018] Furthermore, for each pixel to be classified, the consistency correlation coefficient between it and the time series reference curve of the local land cover category is calculated, and the consistency correlation coefficient corresponding to each land cover category is used as the consistency correlation feature of the pixel to be classified.

[0019] Furthermore, in step S4, the consistency-related features under different polarization modes are used to construct a classification feature vector according to a preset combination method. The combination method includes at least one of the following: consistency-related features using a single polarization mode, a combination of consistency-related features under different polarization modes, and a combination of consistency-related features constructed based on a combination of backscattering coefficients under different polarization modes. The combination of backscattering coefficients includes the product of backscattering coefficients under different polarization modes.

[0020] Furthermore, in step S5, the crop classification model is constructed using a machine learning classification model, and the model is trained using the classification feature vectors corresponding to the training sample set.

[0021] Furthermore, the machine learning classification model is a random forest model, a lightweight gradient booster model, or other ensemble learning models.

[0022] Furthermore, before training the crop classification model, the sample set is divided into a training sample set and a validation sample set. The training sample set is used to construct a standard time-series reference curve and train the crop classification model, while the validation sample set is used to verify the classification accuracy of the trained crop classification model.

[0023] A time-series radar crop classification system based on consistency coefficients, comprising:

[0024] The data acquisition module is used to acquire temporal polarimetric synthetic aperture radar images covering the target area, and to preprocess the temporal polarimetric synthetic aperture radar images to obtain temporal backscattering coefficient data under different polarization modes.

[0025] The reference curve construction module is used to construct a training sample set based on the time-series backscattering coefficient data, and construct standard time-series reference curves for each land cover category under each polarization mode based on the time-series backscattering curves corresponding to each land cover category in the training sample set.

[0026] The consistency correlation feature construction module is used to perform consistency correlation analysis on the temporal backscattering curves of the pixels to be classified under each polarization mode and the standard temporal reference curves corresponding to each land cover category, to obtain the consistency correlation coefficients of the pixels to be classified for each land cover category, and to construct a classification feature vector based on the consistency correlation coefficients obtained under different polarization modes.

[0027] The model training module is used to train a crop classification model using the classification feature vectors corresponding to the training sample set.

[0028] The classification prediction module is used to predict the classification feature vectors corresponding to the pixels to be classified in the target area using the trained crop classification model, so as to obtain the crop classification results in the target area.

[0029] The present invention has the following advantages:

[0030] This invention is the first to introduce the Consistency Correlation Coefficient (CCC) into the field of crop classification using temporal SAR imagery. By calculating the consistency between the temporal curve of the pixel to be classified and the standard reference curve, a novel classification feature with clear physical meaning and strong separability is constructed. This feature can effectively characterize the temporal stability and variation patterns of ground object scattering characteristics, fundamentally avoiding the dependence on prior phenological knowledge in traditional methods and reducing the number of feature parameters for classification, significantly improving the automation level and cross-regional applicability of the method. This invention fully explores the complementary advantages of multi-polarization information, designing various CCC feature combination schemes, such as VV_CCC, VH_CCC, and VV*VH_CCC (VV represents the polarization mode of SAR electromagnetic wave vertical transmission and vertical reception, and VH represents the polarization mode of SAR electromagnetic wave vertical transmission and horizontal reception). An ensemble learning classifier is used for automatic feature selection and classification decision-making, which can effectively distinguish different crop categories (such as rice, soybeans, and ramie) with highly overlapping original backscattering coefficient values ​​but subtle differences in temporal patterns, significantly improving the precision and robustness of crop classification. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of the overall process for a time-series radar crop classification method based on consistency correlation coefficient.

[0033] Figure 2 This is a schematic diagram of the standard time series reference curves for various land features constructed in the embodiments of the present invention;

[0034] Figure 3 This is a diagram showing the classification results of multi-category crops in an embodiment of the present invention;

[0035] Figure 4 This is a diagram showing the classification results of rice as a crop in an embodiment of the present invention.

[0036] Figure 5 This is a diagram showing the results of the soybean binary crop classification in an embodiment of the present invention;

[0037] Figure 6 This is a diagram showing the classification results of ramie as a crop in an embodiment of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0039] Example 1:

[0040] like Figure 1 As shown, this embodiment provides a crop classification method based on the consistency correlation coefficient of radar time-series backscattering coefficients, including the following steps:

[0041] S1. Obtain a set of polarimetric SAR time-series images covering the target area, and preprocess all images to obtain backscattering coefficient images for different polarization modes in each time phase.

[0042] S2. Based on the preprocessed time-series images, a training sample set containing multiple land cover types is constructed, and the pixel time-series curves of each land cover type in the training sample set are used to construct the standard time-series reference curve for that land cover type.

[0043] S3. Based on the standard time series curve, calculate the consistency correlation coefficient between the SAR time series observation value and the standard time series curve of various land features for each pixel, and combine them according to different polarization modes;

[0044] S4. Use the consistency correlation coefficients under the different polarization modes mentioned above as input features, combine them with the training sample set, train and construct a crop classification model.

[0045] S5. Use the trained crop classification model to predict the entire target area map, obtain the crop classification result map, and evaluate the classification accuracy.

[0046] In this embodiment of the invention, in S1, the preprocessing includes orbit correction, radiometric correction, multi-view, filtering, geocoding, and further converts the backscattering coefficient from decibel (dB) form to a linear scale for subsequent time series analysis.

[0047] In this embodiment of the invention, in S2, constructing a standard time-series reference curve specifically involves: extracting the time-series curves of all training pixels in the corresponding category under VV and VH polarization from the training sample set according to preset multiple land cover categories; for each land cover category and each polarization mode, averaging the time-series curves of all training pixels over time to reduce the influence of random noise and individual differences, thereby obtaining the standard time-series reference curve of that land cover category under that polarization mode, which can reflect the overall time-series variation law of that category. The Sentinel-1 satellite IW mode dual-polarization SAR data includes VV polarization (vertical transmission, vertical reception) and VH polarization (vertical transmission, horizontal reception).

[0048] In this embodiment of the invention, in step S3, the consistency correlation coefficient (CCC) between the SAR time series observation value and the standard time series reference curves of various land features is calculated pixel by pixel. The calculation formula is as follows:

[0049]

[0050]

[0051] In the formula, The Pearson correlation coefficient; The correlation coefficient is the consistency coefficient. Indicates the pixel to be classified in the th order. Backscattering coefficients for each phase, This indicates the standard time series reference curve for the corresponding land cover category in the [number]th [year]. Backscattering coefficients for each phase; and These represent the mean values ​​of the temporal backscattering coefficient of the pixel to be classified and the backscattering coefficient of the standard temporal reference curve, respectively. and These represent the standard deviations of the temporal backscattering coefficient of the pixel to be classified and the backscattering coefficient of the standard temporal reference curve, respectively. Indicates the phase number, This represents the total number of time phases observed in the time series.

[0052] In one implementation, a combined consistency correlation feature can also be constructed based on the combined time-series curves of backscattering coefficients of different polarization modes; for example, the backscattering coefficients of different polarization modes are multiplied one time-phase by one to obtain the combined time-series backscattering curve, and the corresponding combined standard time-series reference curve and combined consistency correlation coefficient are constructed according to the above method.

[0053] In this embodiment of the invention, in S4, the crop classification model is constructed using ensemble learning algorithms such as Random Forest or Lightweight Gradient Boosting Machine (LightGBM); during model training, grid search or early stopping based on the validation set is used to optimize the selection of key hyperparameters of the model (such as the number of decision trees) to improve the generalization performance of the model.

[0054] In this embodiment of the invention, step S4 further includes an evaluation step of the model classification accuracy: dividing the training sample set into a training subset and a validation subset according to a preset ratio; the training subset is used to construct a standard time series reference curve and train the classification model; the validation subset is only used to verify the accuracy of the trained model and does not participate in the construction of the reference curve and the training process of the model, so as to ensure the objectivity of the evaluation results.

[0055] Example 2:

[0056] The present invention will now be described with reference to specific embodiments.

[0057] This invention utilizes 20 sets of Sentinel-1 satellite IW mode dual-polarization (VV and VH) SAR data covering the target study area in 2024, totaling 20 sets (20 sets for each polarization). The data originates from the European Space Agency (ESA) official scientific data hub. The acquired SAR data was preprocessed using SNAP (Sentinel Application Platform) 10.0 software. Preprocessing included orbit correction, radiometric correction, multi-look processing (resampling to 10 m resolution), refined Lee filtering (5×5 window), geocoding (UTM / WGS84 projection), and converting the backscattering coefficients from decibels (dB) to a linear scale. The final result is a time-series image cube of VV and VH polarization linear backscattering coefficients for the study area in 2024.

[0058] Based on field survey data, a sample bank of 10 land cover types was established in the study area: early rice, mid-season rice, soybean, corn, ramie, forage, lotus pond, water body, forest, and buildings. All field samples were randomly divided into training subset and validation subset at a ratio of 7:3.

[0059] Standard time-series reference curves are constructed using only a subset of training samples: For each land cover class, the temporal linear backscattering curves of all training pixels under VV and VH polarizations are extracted, and then averaged time-by-time to obtain the standard time-series reference curve for that land cover class under that polarization mode. The constructed standard time-series reference curves for each land cover class are shown below. Figure 2 As shown. By Figure 2 It can be clearly seen that although the standard time series reference curves of various land features have some overlap and intersection, they have significantly different characteristics in the overall time series change pattern, which provides a physical basis for subsequent classification based on CCC.

[0060] For each pixel within the study area, the consistency correlation coefficient (CCC) is calculated between its full-time backscatter observation vectors with VV and VH polarizations and the standard time-series reference curves of the aforementioned 10 land cover types. The formula for calculating CCC is:

[0061]

[0062]

[0063] In the formula, The Pearson correlation coefficient; The correlation coefficient is the consistency coefficient. Indicates the pixel to be classified in the th order. Backscattering coefficients for each phase, This indicates the standard time series reference curve for the corresponding land cover category in the [number]th [year]. Backscattering coefficients for each phase; and These represent the mean values ​​of the temporal backscattering coefficient of the pixel to be classified and the backscattering coefficient of the standard temporal reference curve, respectively. and These represent the standard deviations of the temporal backscattering coefficient of the pixel to be classified and the backscattering coefficient of the standard temporal reference curve, respectively. Indicates the phase number, This represents the total number of time phases observed in the time series.

[0064] Iterate through all pixels and all categories to generate 10 single-class feature maps of consistency correlation coefficients (VV_CCC) and 10 single-class feature maps of consistency correlation coefficients (VH_CCC) under the VV polarization scheme. To fully utilize polarization information, four feature combination schemes were designed: Scheme A uses only VV_CCC (10-dimensional), Scheme B uses only VH_CCC (10-dimensional), Scheme C concatenates VV_CCC and VH_CCC (20-dimensional), and Scheme D adds the CCC feature of the VV*VH product term to Scheme C (30-dimensional in total). These feature parameters are used as input features, and corresponding feature vectors are extracted from the training subset sample locations for subsequent model training.

[0065] This embodiment performs two types of classification tasks: multi-class classification, merging 10 categories into 4 major categories: rice, soybean, ramie, and others; and binary classification, constructing binary classification models for "target crop vs. other land cover" for each crop (rice, soybean, ramie, etc.). Random Forest (RF) and Lightweight Gradient Boosting Machine (LightGBM) are used as classifiers, respectively. During model training, a grid search method is used to optimize the number of decision trees within the range of 100-1500 (step size 100), selecting the optimal parameters based on the overall accuracy (OA) of the validation set. The validation subset is used only for accuracy verification and is not included in reference curve construction or model training to ensure the objectivity of the evaluation.

[0066] The trained optimal model was used to perform block prediction (block size 1024*1024 pixels) on the CCC feature map of the entire study area, yielding crop classification results. The multi-classification results are as follows: Figure 3 As shown, rice has the widest planting area in the entire region, while other crops have smaller planting areas. Table 1 shows that the classification accuracy for many crops exceeds 94%, and the recall rates for rice, ramie, and other crop categories exceed 96%. The binary classification results for rice are shown below. Figure 4 As shown, the results of the soybean binary classification are as follows: Figure 5 As shown, the binary classification results for ramie are as follows: Figure 6As shown in Table 2, the classification accuracy for the three crop categories reached an overall accuracy of over 96%, with soybeans and ramie achieving an overall classification accuracy of over 99% and a Kappa coefficient exceeding 0.92. In summary, the time-series radar crop classification method based on consistency coefficient proposed in this invention provides an efficient and transferable technical solution for remote sensing monitoring of crop planting areas.

[0067] Table 1 - Accuracy of Multi-Classification of Crops

[0068] crop P (accuracy) R (Recall Rate) F1 FR rice 94.68% 96.64% 95.65% 2.94% soybeans 94.02% 92.73% 93.37% 0.22% Ramie 98.40% 97.78% 98.09% 0.04% other 97.96% 96.86% 97.41% 2.88%

[0069] Table 2 - Accuracy of Binary Classification of Crops

[0070] crop OA Kappa F1 FR rice 96.71% 92.81% 95.36% 3.15% soybeans 99.51% 92.80% 93.05% 0.18% Ramie 99.92% 98.19% 98.24% 0.02%

[0071] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0072] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A time-series radar crop classification method based on consistency coefficient, characterized in that, Includes the following steps: S1. Acquire temporal polarimetric synthetic aperture radar (TPSAR) images covering the target area, and preprocess the temporal polarimetric synthetic aperture radar (TPSAR) images to obtain temporal backscattering coefficient data under different polarization modes. S2. Construct a training sample set based on the time-series backscattering coefficient data, and construct a standard time-series reference curve for each land cover category under each polarization mode based on the time-series backscattering curves corresponding to each land cover category in the training sample set. S3. For the pixel to be classified, perform consistency correlation analysis between its time-series backscattering curve under each polarization mode and the standard time-series reference curve corresponding to each land cover category to obtain the consistency correlation coefficient of the pixel to be classified corresponding to each land cover category. S4. Construct the classification feature vector of the pixel to be classified based on the consistency correlation coefficient obtained under different polarization modes; S5. Train a crop classification model using the classification feature vectors corresponding to the training sample set; S6. Using the trained crop classification model, predict the classification feature vectors corresponding to the pixels to be classified in the target area to obtain the crop classification results for the target area.

2. The time-series radar crop classification based on consistency coefficient according to claim 1, characterized in that, In step S1, the preprocessing includes orbit correction, radiometric correction, multi-view processing, filtering, geocoding, and backscattering coefficient linearization to obtain time-series linear backscattering coefficient data under different polarization modes.

3. The time-series radar crop classification based on consistency coefficient according to claim 1, characterized in that, Step S2 includes: Based on the preset land cover categories, the temporal backscattering curves of training pixels of each category under each polarization mode are extracted from the training sample set; For each land cover category, under each polarization mode, the backscattering of training pixels of the same category is averaged in each time phase. That is, the temporal backscattering curve of training pixels for each land cover category is averaged phase by phase, and the average values ​​of each phase are used to construct a standard time-series reference curve. The standard time-series reference curve corresponding to this land cover category is obtained.

4. The time-series radar crop classification based on consistency coefficient according to claim 1, characterized in that, In step S3, the consistency correlation analysis is performed using the consistency correlation coefficient, which comprehensively reflects the correlation and consistency between the temporal backscattering curve of the pixel to be classified and the standard temporal reference curve.

5. The time-series radar crop classification based on consistency coefficient according to claim 4, characterized in that, For each pixel to be classified, the consistency correlation coefficient between it and the time series reference curve of the local land cover category is calculated, and the consistency correlation coefficient corresponding to each land cover category is used as the consistency correlation feature of the pixel to be classified.

6. The time-series radar crop classification based on consistency coefficient according to claim 1, characterized in that, In step S4, the consistency-related features under different polarization modes are combined in a preset manner to construct a classification feature vector. The combination manner includes at least one of the following: consistency-related features using a single polarization mode, a combination of consistency-related features under different polarization modes, and a combination of consistency-related features constructed based on a combination of backscattering coefficients under different polarization modes. The combination of backscattering coefficients includes the product of backscattering coefficients under different polarization modes.

7. The time-series radar crop classification based on consistency coefficient according to claim 1, characterized in that, In step S5, the crop classification model is constructed using a machine learning classification model, and the model is trained using the classification feature vectors corresponding to the training sample set.

8. The time-series radar crop classification based on consistency coefficient according to claim 7, characterized in that, The machine learning classification model is a random forest model, a lightweight gradient booster model, or other ensemble learning models.

9. The time-series radar crop classification based on consistency coefficient according to claim 1, characterized in that, Before training the crop classification model, the sample set is divided into a training sample set and a validation sample set. The training sample set is used to construct a standard time series reference curve and train the crop classification model, while the validation sample set is used to verify the classification accuracy of the trained crop classification model.

10. A time-series radar crop classification system based on consistency coefficients, characterized in that, include: The data acquisition module is used to acquire temporal polarimetric synthetic aperture radar images covering the target area, and to preprocess the temporal polarimetric synthetic aperture radar images to obtain temporal backscattering coefficient data under different polarization modes. The reference curve construction module is used to construct a training sample set based on the time-series backscattering coefficient data, and construct standard time-series reference curves for each land cover category under each polarization mode based on the time-series backscattering curves corresponding to each land cover category in the training sample set. The consistency correlation feature construction module is used to perform consistency correlation analysis on the temporal backscattering curves of the pixels to be classified under each polarization mode and the standard temporal reference curves corresponding to each land cover category, to obtain the consistency correlation coefficients of the pixels to be classified for each land cover category, and to construct a classification feature vector based on the consistency correlation coefficients obtained under different polarization modes. The model training module is used to train a crop classification model using the classification feature vectors corresponding to the training sample set. The classification prediction module is used to predict the classification feature vectors corresponding to the pixels to be classified in the target area using the trained crop classification model, so as to obtain the crop classification results in the target area.

Images