Crop planting area rapid extraction method and system based on radar feature matching

Abstract

The application relates to the technical field of pattern recognition, and provides a crop planting area rapid extraction method and system based on radar feature matching, a computer readable storage medium and an electronic device, wherein the crop planting area rapid extraction method based on radar feature matching comprises the following steps: performing feature screening on full-time sequence backscattering coefficients of radar data of a training area to obtain an optimal feature subset; the full-time sequence backscattering coefficients comprise backscattering coefficients of multiple phases in a growth cycle of crops; time alignment is performed on the optimal feature subset according to the phenological phase of the crops, and a crop recognition model is trained according to the time-aligned optimal feature subset; spatial feature factor analysis is performed on a training area and a target area according to a spatial generalization influence factor of the training area and the target area to obtain a spatial feature-matched target area; and the crop planting area in the spatial feature-matched target area is extracted based on the crop recognition model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of pattern recognition technology, and in particular to a crop planting area rapid extraction method and system based on radar feature matching, a computer readable storage medium and an electronic device. BACKGROUND

[0002] Crop planting area distribution is the basis for crop growth monitoring and yield estimation.

[0003] In related technologies, methods for obtaining crop planting area distribution include statistical methods, remote sensing information extraction methods, and machine learning supervised classification methods. However, the statistical method has a long statistical cycle and high cost. The remote sensing information extraction method based on optical features is susceptible to weather conditions and difficult to obtain high-quality data during the crop growth cycle. Although the machine learning supervised classification method can identify crops in specific time phases and specific regions, it relies on training samples. When the planting area of crops in different times and different spaces needs to be extracted, the model needs to be retrained with training samples from the region, which is time-consuming and labor-intensive, and the efficiency of planting area extraction is not high.

[0004] Therefore, there is a need to provide an improved technical solution to address the above-mentioned deficiencies in the prior art. SUMMARY

[0005] The present application aims to provide a crop planting area rapid extraction method and system based on radar feature matching, a computer readable storage medium and an electronic device to solve or alleviate the problems in the prior art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] The present application provides a crop planting area rapid extraction method based on radar feature matching, comprising:

[0008] Feature screening is performed on the full-time backscattering coefficients of the radar data of the obtained training area to obtain an optimal feature subset; wherein the full-time backscattering coefficients include backscattering coefficients of multiple time phases within the growth cycle of the crop;

[0009] The optimal feature subset is time-aligned according to the phenological period of the crop, and a crop recognition model is trained according to the time-aligned optimal feature subset;

[0010] According to the spatial generalization influence factor of the training area and the target area, spatial feature factor analysis is performed on the training area and the target area to obtain a spatial feature matched target area; and the crop planting area in the spatial feature matched target area is extracted based on the crop recognition model.

[0011] Optionally, the full-time backscattering coefficients of the radar data of the training area are feature-screened to obtain an optimal feature subset, specifically:

[0012] According to the full-time backscattering coefficients, a time-series cumulative recognition accuracy of the crops is determined;

[0013] According to the time-series cumulative recognition accuracy of the crops, a crop optimal recognition time period of the training area is determined;

[0014] According to the crop optimal recognition time period of the training area, the full-time backscattering coefficients are feature-screened to obtain the optimal feature subset.

[0015] Optionally, the full-time backscattering coefficients of the radar data of the training area are feature-screened to obtain an optimal feature subset, specifically:

[0016] According to the polarization mode of the radar data, an importance score of the backscattering coefficient of each time phase in the full-time backscattering coefficient is obtained, to obtain a first score, wherein the first score represents a contribution value of the backscattering coefficient of each time phase to the recognition accuracy of the crop recognition model;

[0017] According to the first score, the backscattering coefficient of each time phase is sorted;

[0018] According to a preset rule, the sorted backscattering coefficients are combined to obtain a plurality of feature sets;

[0019] The recognition accuracy of each feature set is calculated;

[0020] According to the cumulative value of the recognition accuracy of the feature set, the optimal feature subset is determined.

[0021] Optionally, the optimal feature subset is time-aligned according to the phenophase of the crops, specifically:

[0022] According to the time corresponding to the late flowering stage in the phenophase of the crops and the pod stage in the phenophase of the crops, a feature synthesis time is determined;

[0023] According to the feature synthesis time, the optimal feature subset is time-aligned.

[0024] Optionally,

[0025] The crop recognition model is trained according to the time-aligned optimal feature subset, specifically:

[0026] The sample data of the training area obtained and the optimal feature subset after time alignment are input into a pre-constructed identification model for training, so as to obtain the crop identification model.

[0027] Optionally, according to the spatial generalization influence factor of the training area and the target area, spatial feature factor analysis is performed on the training area and the target area, so as to obtain a target area matched in spatial features.

[0028] According to the spatial generalization influence factor of the training area and the target area, the ground surface landscape type of the training area and the ground surface landscape type of the target area are determined respectively.

[0029] According to the ground surface landscape type of the training area and the ground surface landscape type of the target area, spatial feature factor analysis is performed on the training area and the target area, so as to obtain a target area matched in spatial features.

[0030] Optionally, the spatial generalization influence factor at least includes a landform type, a plot size, a slope, and a plot fragmentation degree.

[0031] Embodiments of the present application also provide a crop planting area rapid extraction system based on radar feature matching, comprising:

[0032] A feature screening unit is configured to perform feature screening on the full-time sequence backscattering coefficients of the radar data of the training area obtained, so as to obtain an optimal feature subset; wherein the full-time sequence backscattering coefficients include backscattering coefficients of multiple phases in the growth cycle of the crop.

[0033] A time alignment unit is configured to perform time alignment on the optimal feature subset according to the phenological phase of the crop, and to train a crop identification model according to the optimal feature subset after time alignment.

[0034] A spatial analysis unit is configured to perform spatial feature factor analysis on the training area and the target area according to a spatial generalization influence factor of the training area and the target area, so as to obtain a target area matched in spatial features; and to extract a crop planting area in the target area matched in spatial features based on the crop identification model.

[0035] Embodiments of the present application also provide a computer readable storage medium having a computer program stored thereon, wherein the computer program is the crop planting area rapid extraction method based on radar feature matching.

[0036] The embodiment of the present application also provides an electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor, and the processor implements the radar feature matching based crop planting area rapid extraction method according to any one of the above embodiments when executing the program.

[0037] Advantages:

[0038] In the present application, the optimal feature subset is obtained by performing feature screening on the full-time backscattering coefficients of the radar data of the acquired training area, wherein the full-time backscattering coefficients include backscattering coefficients at multiple time phases in the growth cycle of crops; the optimal feature subset is time-aligned according to the phenological stages of crops, and a crop recognition model is trained according to the time-aligned optimal feature subset; the spatial feature factor analysis of the training area and the target area is performed according to the spatial generalization influence factor of the training area and the target area, and the spatial feature matching target area is obtained; the crop planting area in the spatial feature matching target area is extracted based on the crop recognition model, so that the crop recognition model is trained based on the sample data of the training area, and the crop recognition model is generalized to the target area in space and time, so that the crop recognition and planting area extraction of the target area do not need to retrain the model, and the crop planting area is quickly extracted, thereby providing accurate data support for growth monitoring and yield estimation.

[0039] The optimal feature subset is time-aligned according to the phenological stages of crops, and then the crop recognition model is trained based on the time-aligned feature subset, so that the model has stable time generalization features, and the time generalization accuracy of the crop recognition model is improved.

[0040] According to the spatial generalization influence factor, the spatial feature factor analysis of the training area and the target area is performed, and the spatial feature matching target area is determined, so that the stability of the spatial generalization of the crop recognition model is ensured, and the crop recognition accuracy of the target area in the spatial generalization is improved. BRIEF DESCRIPTION OF DRAWINGS

[0041] The drawings accompanying the specification of the present application serve to provide a further understanding of the present application, the illustrative embodiments of the present application and the explanations thereof serve to explain the present application, and do not constitute improper limitations on the present application. Among them:

[0042] Figure 1 A flowchart of a radar feature matching based crop planting area rapid extraction method according to some embodiments of the present application is shown in the figure;

[0043] Figure 2 A technical flowchart of a radar feature matching based crop planting area rapid extraction method according to some embodiments of the present application is shown in the figure;

[0044] Figure 3 a time-series distribution map of radar data for the first period and the second period;

[0045] Figure 4 a crop time-series cumulative identification precision curve obtained according to the total time-series backscattering coefficient;

[0046] Figure 5 a backscattering coefficient importance score provided according to some embodiments of the present application for different time phases;

[0047] Figure 6 a cumulative identification precision curve of different feature sets provided according to some embodiments of the present application;

[0048] Figure 7 a comparison diagram of identification precision of different time window scales provided according to some embodiments of the present application;

[0049] Figure 8 a crop planting area extraction result diagram provided according to some embodiments of the present application;

[0050] Figure 9 a structure diagram of a crop planting area rapid extraction system based on radar feature matching provided according to some embodiments of the present application;

[0051] Figure 10 a structure diagram of an electronic device provided according to some embodiments of the present application;

[0052] Figure 11 a hardware structure of an electronic device provided according to some embodiments of the present application. DETAILED DESCRIPTION

[0053] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. Each example is provided by way of explanation of the present application rather than limitation of the present application. It will be apparent to those skilled in the art that modifications and variations can be made in the present application without departing from the scope or spirit of the present application. For example, features shown or described as part of one embodiment can be used in another embodiment to produce yet another embodiment. Therefore, it is intended that the present application include such modifications and variations as fall within the scope of the appended claims and their equivalents.

[0054] Exemplary method

[0055] Figure 1 a flow diagram of a crop planting area rapid extraction method based on radar feature matching provided according to some embodiments of the present application; Figure 2 a technical flowchart of a crop planting area rapid extraction method based on radar feature matching provided according to some embodiments of the present application; asFigure 1 、 Figure 2 The crop planting area rapid extraction method based on radar feature matching includes the following steps as shown in FIG. 1:

[0056] In step S101, the feature screening is performed on the full-time sequence backscattering coefficients of the radar data of the acquired training area to obtain an optimal feature subset; wherein the full-time sequence backscattering coefficients include backscattering coefficients at multiple time phases in the growth cycle of the crops.

[0057] In the embodiments of the present application, the crops can include cereal crops and other crops, wherein the cereal crops can be winter wheat, rice, etc., and the other crops can be rapeseed, etc. It should be noted that rapeseed and winter wheat are two common overwintering crops, which have approximately the same growth cycle and optical characteristics, increasing the difficulty of identifying the extraction of the winter wheat and rapeseed planting areas by remote sensing images. Herein, the extraction of the rapeseed planting area is taken as an example to describe the technical solutions of the embodiments of the present application.

[0058] In the embodiments of the present application, the radar data (Synthetic Aperture Radar, SAR, also referred to as radar image) can be acquired from radar sensors carried by satellites or from sensors on other platforms (such as airplanes). Herein, the technical solutions are described by taking the GRD (Ground Range Detected) data of the Sentinel-1 satellite as an example. The Sentinel-1 GRD data is acquired from the GEE (Google Earth Engine) cloud platform, and is imaged in the IW (Interferometric Wideswath) mode with an azimuthal resolution of 22 m and a range resolution of 20 m.

[0059] In the embodiments of the present application, the backscattering coefficients in the radar data are used as the features for identifying different crops to generate the training data of the crop identification model, wherein the backscattering coefficients can be acquired in a single polarization mode, a multi-polarization mode or a full-polarization mode, or the backscattering coefficients obtained by multiple polarization modes can be combined to obtain a candidate data set of the training data. It can be understood that polarization refers to the vibration direction of the radio wave electromagnetic field used by the sensor when transmitting signals to the ground, and H represents horizontal polarization and V represents vertical polarization. In the embodiments, the Sentinel-1 GRD (Ground Range Detected) data has VV and VH dual polarization modes, that is, each Sentinel-1 GRD (Ground Range Detected) data image contains VV backscattering coefficients and VH backscattering coefficients.

[0060] In the application scenario of the embodiment, the acquired Sentinel-1 GRD data is sequentially processed as follows: 5-view processing, thermal noise removal, polarization scaling, and terrain correction to obtain processed backscatter coefficients, and then the backscatter coefficients obtained by processing are interpolated to obtain incident angle angle data to improve the resolution thereof, and the incident angle angle data is corrected in terms of radar incident angle; finally, a 3*3 window size RefinedLee filter is used to spatially filter the corrected incident angle angle data to obtain backscatter coefficients for identifying different crop characteristics. Here, the backscatter coefficients for identifying different crop characteristics are obtained by preprocessing the Sentinel-1 GRD data, and the spatial resolution of the preprocessed backscatter coefficients is 10 m, which improves the quality of the original radar data and reduces the influence of speckle noise in the radar data.

[0061] In the embodiment, the full-time backscatter coefficients include backscatter coefficients of multiple time phases in the growth period of crops. The full-time backscatter coefficients can be time-series radar data of the growth period of overwintering crops in a certain period (such as a year), or can be time-series radar data of the growth period of overwintering crops in multiple periods. Taking A City as an example, the main overwintering crops planted in A City are winter wheat and winter rape, and the time-series radar data of the growth period of overwintering crops in the first period and the second period can be selected as example data of the backscatter coefficients of multiple time phases, where the first period can be 2018, the second period can be 2019, and the crop growth period in each period is from September to the following July. Therefore, the full-time backscatter coefficients include radar data at different time points in the first period and the second period, and each time point corresponds to a different time phase in the first period or the second period. Specifically, for the acquired Sentinel-1 GRD (Ground Range Detected) data, the first period has 46 different time points, and the VV backscatter coefficients and the VH backscatter coefficients during the growth period of crops in the first period have 46 images respectively, i.e., 46 time-phase radar images are included. The second period has 47 different time points, and the VV backscatter coefficients and the VH backscatter coefficients during the growth period of crops in the second period have 47 images respectively. The time-phase distribution of the radar data in the two periods is shown in FIG. 2. Figure 3

[0062] In some optional embodiments, the full-time backscatter coefficients of the acquired radar data of the training area are subjected to feature screening to obtain an optimal feature subset, specifically as follows: determining the time-series cumulative recognition accuracy of crops according to the full-time backscatter coefficients; determining the optimal recognition time period of crops in the training area according to the time-series cumulative recognition accuracy of crops; and performing feature screening on the full-time backscatter coefficients according to the optimal recognition time period of crops in the training area to obtain an optimal feature subset.

[0063] ​In a specific application, first, the time series cumulative recognition accuracy of crops is determined according to the full-time backscattering coefficient, including: the full-time backscattering coefficient includes backscattering coefficients of multiple time phases in the growth cycle of crops, and each time phase backscattering coefficient includes two parts of VV backscattering coefficient and VH backscattering coefficient. Before determining the time series cumulative recognition accuracy of crops, the VV backscattering coefficient and the VH backscattering coefficient are combined to obtain the combined backscattering coefficient of each time phase, and then the image corresponding to the combined backscattering coefficient is accumulated in time sequence and input into the preset crop recognition model for pre-training, and the time series cumulative recognition accuracy of crops is calculated.

[0064] Then, the time series cumulative recognition accuracy curve of crops is obtained according to the time series cumulative recognition accuracy of crops, and the optimal recognition time period of the training area of crops is determined by analyzing the time series cumulative recognition accuracy curve of crops. Figure 4 The time series cumulative recognition accuracy curve of crops obtained according to the full-time backscattering coefficient; as shown in Figure 4 As shown in the figure, taking the time series cumulative recognition accuracy of rape in the first period as an example, with the accumulation of time series backscattering coefficient, the cumulative recognition accuracy of rape gradually increases, and reaches 0.8 cumulative recognition accuracy before April; during April, the recognition accuracy of rape is improved at a higher rate, and the cumulative recognition accuracy is close to 0.9; after May, the time series backscattering coefficient is continuously accumulated and input into the crop recognition model for pre-training, at this time, the cumulative recognition accuracy curve of rape is relatively stable. According to the distribution of the curve, it can be determined that for the full sequence backscattering coefficient of the first period, the backscattering coefficient before May can meet the accuracy requirement of rape recognition. Further analysis of the crop recognition model trained by the backscattering coefficient of each time point can show that the recognition accuracy of the crop recognition model trained by the backscattering coefficient of three time points in April is higher than that of other time points, and it can be determined that for the training data of the first period, April is the optimal recognition time period of rape.

[0065] Finally, based on the optimal crop recognition time period in the training area, feature screening was performed on the full time series backscatter coefficients to obtain the optimal feature subset. For the rapeseed backscatter coefficients in the first period, the aforementioned analysis indicates that April is the optimal recognition time period. Therefore, feature screening was performed on the full time series backscatter coefficients. Specifically, based on the backscatter coefficients corresponding to different time points, the backscatter coefficients for April were selected to obtain the optimal feature subset. This optimal feature subset includes backscatter coefficients within the optimal crop recognition time period, and each backscatter coefficient includes data for both VV and VH polarization modes.

[0066] In other optional embodiments, feature screening is performed on the full-time series backscatter coefficients of the radar data acquired in the training area to obtain an optimal feature subset, specifically: according to the polarization mode of the radar data, the importance of the backscatter coefficients of each phase in the full-time series backscatter coefficients is scored to obtain a first score, wherein the first score represents the contribution value of the backscatter coefficient of each phase to the recognition accuracy of the crop recognition model; according to the first score, the backscatter coefficients of each phase are sorted; according to the preset rules, the sorted backscatter coefficients are combined to obtain multiple groups of feature sets; the recognition accuracy of each group of feature sets is calculated; and according to the cumulative value of the recognition accuracy of the feature sets, the optimal feature subset is determined.

[0067] In an embodiment of the present application, the backscatter coefficient of each phase in the full time series backscatter coefficient is divided into two parts, namely, the VV backscatter coefficient and the VH backscatter coefficient, according to the polarization mode of the radar data. The VV backscatter coefficient and the VH backscatter coefficient of each phase are respectively input into the crop recognition model for pre-training, and the feature importance scores (first scores) of different phases and different polarization modes are calculated, wherein the feature importance score represents the contribution value of the backscatter coefficient of each phase to the recognition accuracy of the crop recognition model. The feature importance score can be implemented using an algorithm of an existing model, such as an importance score based on a random forest classification model, or it can be implemented based on other importance score algorithms, and the present application does not impose any restrictions on this. Figure 5As shown, the VH backscattering coefficients of the last period of flowering of rape (early April) and the development period of silique (mid and late April) have the highest importance score, indicating that the backscattering coefficients of this phase and polarization mode have the greatest contribution to the recognition accuracy of the crop recognition model. Then, according to the size of the importance score, the backscattering coefficients of each phase and each polarization mode are ranked from high to low to obtain the ranking result. Further, the backscattering coefficients after ranking are combined one by one according to the importance score from high to low to obtain a plurality of feature sets. Finally, the plurality of feature sets are input into the crop recognition model for pre-training, the recognition accuracy of each feature set is calculated, and it is judged whether the cumulative value of the recognition accuracy is greater than the preset threshold. If yes, the feature set with the cumulative value of the recognition accuracy greater than the preset threshold is taken as the optimal feature subset.

[0068] Figure 6 The recognition accuracy cumulative curve of different feature sets provided according to some embodiments of the present application is shown in the figure. Figure 6 As shown, the VV backscattering coefficients and VH backscattering coefficients of the first period of crop growth have 46 images, each image is input into the crop recognition model for pre-training, the importance score corresponding to each image is calculated, and the importance score is ranked. As can be seen from the figure, the VH backscattering coefficient numbered 20180426 has the highest importance score, the VH backscattering coefficient numbered 20180414 has the second highest importance score, and so on. The recognition accuracy of the first 6 numbered backscattering coefficients reaches more than 0.85, and when the recognition accuracy reaches 0.89, the recognition accuracy cumulative curve tends to be stable, and the contribution value of the recognition accuracy of the crop recognition model by continuing to increase the training data is small. Therefore, according to the analysis result of the recognition accuracy cumulative curve of the feature set, 14 images are selected from the 46 time series images (46 phase backscattering coefficients) to obtain the optimal feature subset.

[0069] In the embodiments of the present application, the full-time backscattering coefficient provides rich information, and the backscattering coefficients of different phases have different contributions to the recognition accuracy of the crop recognition model. Therefore, using the optimal feature subset as the training data of the crop recognition model can select effective features (backscattering coefficients) for recognizing crops, reduce the amount of data involved in operation, and improve the crop recognition efficiency.

[0070] In step S102, the optimal feature subset is time-aligned according to the phenological phase of the crop, and the crop recognition model is trained according to the time-aligned optimal feature subset.

[0071] In some optional embodiments, the optimal feature subsets are time-aligned according to the phenological period of the crop, specifically: the feature synthesis time is determined according to the time corresponding to the end of the flowering period in the phenological period of the crop and the silique period in the phenological period of the crop; and the optimal feature subsets are time-aligned according to the feature synthesis time.

[0072] When implementing it specifically, Figure 3 As shown in the figure, the imaging times of radar data acquired by the sensor during different periods cannot be fully aligned. That is, the imaging times of the radar data in the training area and the radar data in the target area are different, which will affect the accuracy of model recognition. In order to obtain stable crop temporal generalization recognition features and realize the migration of crop recognition models across different time periods, the late flowering and silique stages of the crop phenological period are used as feature synthesis times. Based on the feature synthesis time, the optimal feature subset is temporally aligned to improve the temporal generalization accuracy of the crop recognition model. Here, temporal generalization refers to using training data obtained from a specific area in a certain period and training the crop recognition model, and then further applying it to different periods for crop identification and planting area extraction.

[0073] In a specific example, the optimal feature subset is time-aligned according to the feature synthesis time, specifically: the optimal feature subset is synthesized according to the maximum synthesis method for the training data within the feature synthesis time to obtain a feature image covering the training area within the time period, and the feature image is time-aligned with the late flowering stage and the silique stage in the phenological period of the crops in the target area to obtain an image that can be used for temporal generalization.

[0074] Due to the large difference in crop growth period in different regions, the radar data acquisition time in different regions cannot be aligned, thereby affecting the accuracy of the crop identification model. In this regard, the traditional time generalization method usually migrates the crop identification model for different periods in the same region. In order to verify the beneficial effect of the time alignment based on the late flowering stage and the silique stage, first, the full-time sequence radar backscattering coefficient, the specific time window synthesized radar data, and the optimal feature subset based on the late flowering stage and the silique stage are input into the crop identification model for pre-training, to obtain the crop identification accuracy corresponding to the three time alignment methods of the full-time sequence radar backscattering coefficient, the specific time window synthesized radar data, and the optimal feature subset, wherein the specific time window can be 15-day time window and 30-day time window respectively. Then, the identification accuracy corresponding to the full-time sequence radar backscattering coefficient, the 15-day time window, the 30-day time window, and the optimal feature subset based on the late flowering stage and the silique stage is divided into five identification accuracy levels, and the identification accuracy range of each level is 0%-60%, 60%-75%, 75%-80%, 80%-90%, and 90%-100% respectively, and is drawn in the form of a segmented graph. Finally, the training area and the target area are divided into a plurality of 10km x 10km grids, and the identification accuracy of each grid is accumulated and drawn into the corresponding identification accuracy level drawing area according to the level to which the identification accuracy belongs, as shown in FIG. 8. Figure 7 As can be seen from the figure, the identification accuracy corresponding to the 15-day time window and the 30-day time window does not have obvious improvement relative to the identification accuracy corresponding to the full-time sequence radar backscattering coefficient, while the identification accuracy corresponding to the optimal feature subset based on the late flowering stage and the silique stage has obvious improvement relative to the identification accuracy corresponding to the full-time sequence radar backscattering coefficient, the 15-day time window, and the 30-day time window. In addition, the late flowering stage and the silique stage are used as the feature synthesis time, and the time window has a certain span (more than 30 days), which ensures the availability of radar data and guarantees the generalization in time.

[0075] In some optional embodiments, the crop identification model is trained according to the optimal feature subset after time alignment, specifically: inputting the sample data of the training area and the optimal feature subset after time alignment into the pre-constructed identification model for training to obtain the crop identification model.

[0076] The sample data of the training area is obtained through ground investigation. The ground investigation is usually carried out in crop planting areas in November of each year, and the ground true planting information is collected by mapping to obtain sample data of the training area in the form of a plot. Here, the plot refers to a naturally divided area for crop planting, such as a rapeseed plot divided by a ridge. Then, the geometric center point of the plot is extracted to obtain the geographic coordinates of the plot. Finally, the ground true planting information is combined with the geographic coordinates of the plot to obtain the sample data. It should be noted that the sample data is the label data used to train the crop recognition model. According to different crops, the sample data can be divided into four types of ground objects (labels), namely rapeseed, wheat, other cultivated land (including summer crops that have not been sown, other crops) and uncultivated land (including forest land, grassland, water body, building and road, etc.).

[0077] In a specific application, the pre-constructed crop recognition model can be any one of the crop classification models in the related art, for example, it can be constructed based on a neural network model, it can also be constructed based on a traditional machine learning classifier, and it can also be constructed based on a random forest model. As long as it can extract features from sample data to obtain a classification model that can be used to identify crops, it will not be described one by one here. In the following, the construction of a crop recognition model based on a random forest model is taken as an example for description.

[0078] The random forest (RF) classification model is a non-parametric machine learning algorithm composed of random decision trees. In the embodiments of the present application, the sample data and the radar data corresponding to the optimal feature subset are input into the random forest classification model for feature extraction. The importance score is used to evaluate the contribution value of each feature in the training data to the recognition accuracy, and the out-of-bag estimation accuracy (oob score) is used to evaluate the validation accuracy of the model. Compared with other machine learning classifiers, the random forest classification model has good accuracy, fast classification speed for large data, can process high-dimensional features and evaluate the importance of each feature for classification, and can generate an internal unbiased estimate of the generalization error. In addition, random forest can use out-of-bag samples to evaluate the model, without the need to set up a separate validation set.

[0079] In step S103, the spatial feature factor analysis of the training area and the target area is performed according to the spatial generalization influence factor of the training area and the target area, and the target area with matched spatial features is obtained. The crop planting area in the target area with matched spatial features is extracted based on the crop recognition model.

[0080] In the embodiments of the present application, the spatial generalization influence factors at least include: landform type, plot size, slope, plot fragmentation. Specifically, the generalization on the spatial scale is affected by plot size, plot fragmentation, landform type, and terrain slope, etc., wherein the landform type includes five types of plain area, plateau area, hilly area, mountain area, and water body, the plot size is represented by plot area, and the plot fragmentation is represented by the perimeter-area ratio of the plot; the slope data is calculated from the elevation data.

[0081] Specifically, the 30m resolution elevation data is downloaded from GEE and the slope data is calculated, and then the downloaded elevation data and the calculated slope data are resampled to obtain the elevation data and the slope data with a spatial resolution of 10m, so that the spatial resolution is unified with the spatial resolution of the radar data. The initial landform data is the 1:4 million national landform data, and the five landform types of plain area, plateau area, hilly area, mountain area, and water body are obtained by masking and vector labeling on the initial landform data. According to the vector data of the training area and the target area, the area of the plot and the perimeter-area ratio of the plot in the training area and the target area can be calculated respectively, so as to determine the plot size and the plot fragmentation.

[0082] In some optional embodiments, according to the spatial generalization influence factors of the training area and the target area, spatial feature factor analysis is performed on the training area and the target area to obtain a target area matched in spatial features, specifically: according to the spatial generalization influence factors of the training area and the target area, the ground surface landscape type of the training area and the ground surface landscape type of the target area are determined respectively; according to the ground surface landscape type of the training area and the ground surface landscape type of the target area, spatial feature factor analysis is performed on the training area and the target area to obtain a target area matched in spatial features.

[0083] Firstly, according to the values of plot size, plot fragmentation, and slope, the spatial generalization influence factors are classified to obtain a spatial generalization influence factor classification table, as shown in Table 1, and Table 1 is as follows:

[0084] Table 1 Spatial generalization influence factor classification table

[0085]

[0086] As can be seen from Table 1, according to the values of plot size, plot fragmentation, and slope, the ground surface landscape of the training area and the target area can be divided into five levels of I, II, III, IV, and V.

[0087] Then, the five landform types of plain area, plateau area, hilly area, mountain area, and water body are combined with the classification of plot size, plot fragmentation, and slope to determine the ground surface landscape type of the training area and the ground surface landscape type of the target area respectively.

[0088] Finally, according to the ground surface landscape type of the training area and the ground surface landscape type of the target area, the training area and the target area are analyzed in terms of spatial feature factors, and the target area with matched spatial features is obtained.

[0089] The analysis of the spatial feature factors of the training area and the target area includes the analysis of the plot size, plot fragmentation, landform type and slope of the training area and the target area respectively, the determination of the matching rules of the plot size, plot fragmentation, landform type and slope, and the determination of the target area with matched spatial features of the training area according to the matching rules.

[0090] Specifically, for the plot size, if the plot sizes of the training area and the target area belong to the same classification, the larger the plot area of the training area, the higher the generalization accuracy of the crop recognition model in the target area; if the plot sizes of the training area and the target area belong to different classifications, the larger the plot area of the target area, the higher the generalization accuracy of the crop recognition model in the target area. Accordingly, to ensure the recognition accuracy of the crop recognition model in the target area, an area with large plot area and complete plots should be selected as the training area of the crop recognition model, and at the same time, the plot area of the target area should be matched with the training area. Preferably, the area with the plot belonging to the third level of plot size is selected as the training area, and as the plot area of the target area increases, the recognition accuracy is continuously improved. At this time, the influence of the plot size of the target area on the accuracy is offset to some extent, so that the training area can use the crop recognition model in more target areas while ensuring the generalization accuracy.

[0091] For the plot fragmentation, it should be noted that the larger the value of the plot fragmentation, the higher the fragmentation degree of the plot. If the plot fragmentation of the training area and the target area belongs to different classifications, the higher the plot fragmentation of the training area and the target area, the lower the crop recognition accuracy; if the plot fragmentation of the training area and the target area belongs to the same classification, the classification I and II show a trend of increasing plot fragmentation and decreasing generalization accuracy; while the generalization accuracy of the classification III, IV and V shows the characteristics of "high in the middle and low at both ends", that is, when the plot fragmentation of the training area and the target area belongs to the classification III (fragmentation value: 0.15-0.2), the crop recognition accuracy is optimal. At this time, selecting the training area and the target area with the plot fragmentation classified as the third level can improve the accuracy of spatial generalization.

[0092] For the landform type, when the target area is mountainous and tableland, the crop recognition accuracy is low, therefore, under the condition that the landform type of the training area is not limited, an area with landform type not belonging to mountainous and tableland should be selected as the target area for spatial generalization.

[0093] For slope, as the slope classification of the training area and the target area is from I to V, the crop recognition accuracy of the target area is reduced regardless of whether the training area and the target area belong to the same slope classification. Therefore, selecting the training area and the target area with the slope classification of I, II and III for spatial generalization can improve the crop recognition accuracy of the target area.

[0094] In summary, the ground surface landscape types of the training area and the target area affect the spatial generalization ability of the crop recognition model. On the one hand, the training area and the target area with small plot area, large fragmentation, large terrain slope and landform type belonging to mountain and platform have poor crop recognition accuracy when spatial generalization is performed. On the other hand, affected by the imaging resolution of the sensor, the radar data obtained in the area with small plot area and large fragmentation is prone to mixed pixels. The backscattering coefficient of the mixed pixels is greatly different from that of the non-mixed pixels, resulting in low crop recognition accuracy in the area with mixed pixels. Therefore, when training the crop recognition model, the present application selects the area with large plot area, small fragmentation, small terrain slope and landform type not belonging to mountain and platform as the training area, and determines the target area matching the ground surface landscape type of the training area according to the above matching rules. It should be noted that the mixed pixel refers to the backscattering coefficient of a pixel in the radar image being affected by the backscattering coefficients of the surrounding pixels, and the value of the mixed pixel is the mixed value obtained by superimposing the backscattering coefficients of the surrounding pixels and the backscattering coefficient of the pixel itself.

[0095] Finally, the crop planting area in the target area matched by the spatial features of the crop recognition model is extracted, specifically: inputting the optimal feature subset obtained based on the flowering late stage and the silique stage as features into the random forest classification model constructed in advance to train the crop recognition model, and performing spatial feature factor analysis from the plot size, plot fragmentation, slope and landform type to obtain the target area matching the training area, then classifying the crops in the matched target area based on the crop recognition model, and extracting the crop planting area based on the crop classification to obtain the planting area distribution map, as shown in Figure 8 .

[0096] In summary, the present application obtains an optimal feature subset by performing feature screening on the full-time backscattering coefficients of the radar data of the training area; the full-time backscattering coefficients include backscattering coefficients at multiple time phases in the growth cycle of the crops; the optimal feature subset is time-aligned according to the phenological stages of the crops, and a crop recognition model is trained according to the time-aligned optimal feature subset; the spatial feature factor analysis is performed on the training area and the target area according to the spatial generalization influence factor of the training area and the target area, and the target area with matched spatial features is obtained; the crop planting area in the target area with matched spatial features is extracted based on the crop recognition model, so that the crop recognition model is trained based on the sample data of the training area, and the crop recognition model is generalized in space and time to the target area, so that the crop recognition and the extraction of the planting area of the target area are realized without retraining the model, and the planting area of the crops is quickly extracted, thereby providing accurate data support for the growth monitoring and yield estimation.

[0097] The optimal feature subset is time-aligned according to the phenological stages of the crops, and the crop recognition model is trained based on the time-aligned feature subset, so that the model has stable time generalization features, and the time generalization accuracy of the crop recognition model is improved.

[0098] According to the spatial generalization influence factor, the spatial feature factor analysis is performed on the training area and the target area, and the target area with matched spatial features is determined, so that the stability of the spatial generalization of the crop recognition model is ensured, and the crop recognition accuracy of the target area in the spatial generalization is improved.

[0099] Exemplary system

[0100] Figure 9 A structure diagram of a crop planting area rapid extraction system based on radar feature matching according to some embodiments of the present application is shown in FIG. 1. Figure 9 As shown in FIG. 1, the crop planting area rapid extraction system based on radar feature matching includes a feature screening unit 901, a time alignment unit 902, and a spatial analysis unit 903.

[0101] The feature screening unit 901 is configured to perform feature screening on the full-time backscattering coefficients of the radar data of the training area to obtain an optimal feature subset; the full-time backscattering coefficients include backscattering coefficients at multiple time phases in the growth cycle of the crops.

[0102] The time alignment unit 902 is configured to time-align the optimal feature subset according to the phenological stages of the crops, and train a crop recognition model according to the time-aligned optimal feature subset.

[0103] The spatial analysis unit 903 is configured to: perform spatial feature factor analysis on the training area and the target area according to the spatial generalization influencing factors of the training area and the target area to obtain a target area with spatial feature matching; and extract the crop planting area within the target area with spatial feature matching based on the crop recognition model.

[0104] The rapid crop planting area extraction system based on radar feature matching provided in the embodiment of the present application can implement the steps and processes of the rapid crop planting area extraction method based on radar feature matching provided in any of the above embodiments, and achieve the same technical effects, which will not be repeated here.

[0105] Exemplary device

[0106] Figure 10 Schematic diagram of the structure of an electronic device according to some embodiments of the present application; Figure 10 As shown, the electronic device includes:

[0107] One or more processors 1001;

[0108] The computer-readable medium may be configured to store one or more programs 1002. When one or more processors 1001 execute the one or more programs 1002, the following steps are implemented:

[0109] The full-time series backscatter coefficients of the radar data acquired in the training area are subjected to feature screening to obtain an optimal feature subset; wherein the full-time series backscatter coefficients include backscatter coefficients of multiple phases within the growth cycle of the crop; the optimal feature subset is time-aligned according to the phenological period of the crop, and a crop recognition model is obtained by training the optimal feature subset after time alignment; according to the spatial generalization influencing factors of the training area and the target area, a spatial feature factor analysis is performed on the training area and the target area to obtain a target area with spatial feature matching; and the crop planting area within the target area with spatial feature matching is extracted based on the crop recognition model.

[0110] Figure 11 The hardware structure of the electronic device provided in some embodiments of the present application is as follows: Figure 11 As shown, the hardware structure of the electronic device may include: a processor 1101 , a communication interface 1102 , a computer-readable medium 1103 and a communication bus 1104 .

[0111] The processor 1101 , the communication interface 1102 , and the computer-readable medium 1103 communicate with each other via a communication bus 1104 .

[0112] Optionally, the communication interface 1102 can be an interface of a communication module, such as an interface of a GSM module.

[0113] The processor 1101 can be specifically configured to:

[0114] The acquired full-time backscattering coefficients of the radar data of the training area are subjected to feature screening to obtain an optimal feature subset; the full-time backscattering coefficients include backscattering coefficients of multiple time phases in a growth cycle of the crops; the optimal feature subset is subjected to time alignment according to phenophase of the crops, and a crop recognition model is trained according to the time-aligned optimal feature subset; spatial feature factor analysis is performed on the training area and a target area according to a spatial generalization influence factor of the training area and the target area to obtain a spatial feature-matched target area; and a crop planting area in the spatial feature-matched target area is extracted based on the crop recognition model.

[0115] The processor can be a general processor, including a central processing unit (CPU), a network processor (NP), etc., and can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a ready programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. The disclosed methods, steps and logic block diagrams in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can be any conventional processor.

[0116] The electronic device of the embodiments of the present application exists in various forms, including but not limited to:

[0117] (1) Mobile communication device: the feature of this kind of device is to have mobile communication function, and to provide voice and data communication as the main target. This kind of terminal includes: smart phone (such as iPhone), multimedia phone, functional phone, and low-end phone, etc.

[0118] (2) Ultra-mobile personal computer device: this kind of device belongs to the category of personal computer, has computing and processing functions, and generally also has mobile Internet characteristics. This kind of terminal includes: PDA, MID and UMPC device, etc., such as iPad.

[0119] (3) Portable entertainment device: this kind of device can display and play multimedia content. This kind of device includes: audio and video player (such as iPod), palm game console, electronic book, and smart toy and portable vehicle navigation device.

[0120] (4) Server: a device providing computing services, the server is composed of a processor, a hard disk, a memory, a system bus, etc., the server is similar to a general computer architecture, but since it needs to provide high-reliable services, it has higher requirements in processing capability, stability, reliability, security, scalability, manageability, etc.

[0121] (5) Other electronic devices with data interaction function.

[0122] It should be noted that, according to the needs of implementation, each component / step described in the embodiments of the present application can be split into more components / steps, or two or more components / steps or part of the operation of the components / steps can be combined into a new component / step, to achieve the purpose of the embodiments of the present application.

[0123] The above method according to the embodiments of the present application can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium such as a CD ROM, a RAM, a floppy disk, a hard disk or an optical disk, or downloaded through a network and originally stored in a remote recording medium or a non-transitory machine storage medium and then stored in a local recording medium, so that the method described herein can be processed by such software on a recording medium using a general computer, a special processor or programmable or special hardware such as an ASIC or an FPGA. It can be understood that the computer, processor, microprocessor controller or programmable hardware includes a storage component (for example, RAM, ROM, flash memory, etc.) that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, processor or hardware, the radar feature matching based crop planting area rapid extraction method described herein is implemented. In addition, when the general computer accesses the code for implementing the method shown herein, the execution of the code will convert the general computer into a special computer for executing the method shown herein.

[0124] Those of ordinary skill in the art can realize that the units and method steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware, or in a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application of the technical solution and the involved constraints. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of the present application.

[0125] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device and system embodiments described above are merely illustrative, wherein the units not described as separate may or may not be physically separated, and the units indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. A person of ordinary skill in the art can understand and implement it without making any creative effort.

[0126] The foregoing description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A method for rapid extraction of crop planting areas based on radar feature matching, characterized in that: include: Performing feature screening on the full-time series backscatter coefficients of the radar data acquired in the training area to obtain an optimal feature subset; wherein the full-time series backscatter coefficients include backscatter coefficients of multiple time phases within the growth cycle of the crop; Time-aligning the optimal feature subset according to the phenological period of the crop, and training a crop recognition model based on the time-aligned optimal feature subset; performing spatial feature factor analysis on the training area and the target area according to the spatial generalization influencing factors of the training area and the target area to obtain a target area with spatial feature matching; and extracting a crop planting area within the target area with spatial feature matching based on the crop recognition model; The spatial generalization influencing factors include at least: landform type, plot size, slope, and plot fragmentation; performing spatial feature factor analysis on the training area and the target area, including: analyzing the plot size, plot fragmentation, landform type, and slope of the training area and the target area respectively, determining a matching rule for the plot size, plot fragmentation, landform type, and slope, and determining a target area that matches the spatial characteristics of the training area based on the matching rule; The time alignment of the optimal feature subset according to the phenological period of the crop is specifically as follows: Determining the characteristic synthesis time according to the time corresponding to the end of flowering in the crop phenological period and the silique stage in the crop phenological period; The optimal feature subset of the training data is synthesized according to the maximum synthesis method within the feature synthesis time to obtain a feature image covering the training area within the time period, and the feature image is aligned with the late flowering and silique stages of the crops in the target area.

2. The method for rapid crop planting area extraction based on radar feature matching according to claim 1 is characterized in that: The feature screening of the full time series backscatter coefficients of the radar data acquired in the training area to obtain the optimal feature subset is specifically: Determining the time series cumulative recognition accuracy of the crop according to the full time series backscatter coefficient; Determining an optimal crop recognition time period in the training area based on the cumulative recognition accuracy of the crop over time series; According to the optimal crop recognition time period in the training area, feature screening is performed on the full time series backscatter coefficients to obtain the optimal feature subset.

3. The method for rapid crop planting area extraction based on radar feature matching according to claim 1 is characterized in that: The feature screening of the full time series backscatter coefficients of the radar data acquired in the training area to obtain the optimal feature subset is specifically: performing an importance score on the backscatter coefficient of each time phase in the full time series of backscatter coefficients according to the polarization mode of the radar data to obtain a first score, wherein the first score represents a contribution value of the backscatter coefficient of each time phase to the recognition accuracy of the crop recognition model; sorting the backscatter coefficients of each phase according to the first score; Combining the sorted backscatter coefficients according to preset rules to obtain multiple sets of feature sets; Calculating the recognition accuracy of each set of feature sets; The optimal feature subset is determined according to the cumulative value of the recognition accuracy of the feature set.

4. The method for rapid crop planting area extraction based on radar feature matching according to claim 1, characterized in that: The crop recognition model is obtained by training the optimal feature subset after time alignment, specifically: The sample data of the training area and the time-aligned optimal feature subset are input into a pre-built recognition model for training to obtain the crop recognition model.

5. The method for rapid crop planting area extraction based on radar feature matching according to claim 1, characterized in that: The spatial feature factor analysis is performed on the training area and the target area according to the spatial generalization influence factors of the training area and the target area to obtain the target area with spatial feature matching, specifically: determining the surface landscape type of the training area and the surface landscape type of the target area respectively according to the spatial generalization influencing factors of the training area and the target area; According to the surface landscape type of the training area and the surface landscape type of the target area, spatial feature factor analysis is performed on the training area and the target area to obtain a target area with matching spatial features.

6. A rapid crop planting area extraction system based on radar feature matching, characterized in that: include: A feature screening unit is configured to perform feature screening on the full-time series backscatter coefficients of the radar data acquired in the training area to obtain an optimal feature subset; wherein the full-time series backscatter coefficients include backscatter coefficients of multiple time phases within the growth cycle of the crop; a time alignment unit configured to: perform time alignment on the optimal feature subset according to the phenological period of the crop, and train a crop recognition model based on the optimal feature subset after time alignment; A spatial analysis unit is configured to: perform spatial feature factor analysis on the training area and the target area according to the spatial generalization influencing factors of the training area and the target area to obtain a target area with spatial feature matching; and extract a crop planting area within the target area with spatial feature matching based on the crop recognition model; The spatial generalization influencing factors include at least: landform type, plot size, slope, and plot fragmentation; performing spatial feature factor analysis on the training area and the target area, including: analyzing the plot size, plot fragmentation, landform type, and slope of the training area and the target area respectively, determining a matching rule for the plot size, plot fragmentation, landform type, and slope, and determining a target area that matches the spatial characteristics of the training area based on the matching rule; The time alignment of the optimal feature subset according to the phenological period of the crop is specifically as follows: Determining the characteristic synthesis time according to the time corresponding to the end of flowering in the crop phenological period and the silique stage in the crop phenological period; The optimal feature subset of the training data is synthesized according to the maximum synthesis method within the feature synthesis time to obtain a feature image covering the training area within the time period, and the feature image is aligned with the late flowering and silique stages of the crops in the target area.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for rapidly extracting crop planting areas based on radar feature matching as described in any one of claims 1 to 5 is implemented.

8. An electronic device, characterized in that: include: A memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for rapidly extracting crop planting areas based on radar feature matching as described in any one of claims 1 to 5 is implemented.

Images