Classification method, device, electronic device, storage medium and product

Abstract

The embodiments of the present application provide a classification method, apparatus, electronic device, storage medium, and product. The method first obtains first, second, and third image data corresponding to a target area within a preset time range. Texture feature data corresponding to a first time unit of the target area is then determined based on the first image data. Multiple spectral feature data and multiple exponential feature data corresponding to a second time unit of the target area are determined based on the second image data. Multiple polarization feature data corresponding to a third time unit of the target area are determined based on the third image data. A feature data set is determined based on the texture feature data, multiple spectral feature data, multiple exponential feature data, and multiple polarization feature data. Finally, the crop type of the target area is determined based on the feature data set and a general classification model. In this way, different image data are converted into corresponding feature data, which are then combined with the classification model to achieve refined classification of crops.

Description

Technical Field

[0001] The present application relates to the field of remote sensing application technology, and is related to but not limited to a classification method, device, electronic device, storage medium and product. Background Art

[0002] Crop classification studies the identification and classification of crops in plain areas, primarily at a resolution of 10 meters or coarser. Traditional crop classification techniques primarily use single-period optical remote sensing imagery and radar imagery to identify and classify crops in plain areas. However, for crops grown in complex mountainous areas and under complex climatic conditions, single-period optical remote sensing imagery is subject to poor accuracy due to the influence of inclement weather. While single-period radar imagery is not affected by inclement weather, its noise can reduce crop classification accuracy.

[0003] Therefore, traditional crop classification technology is difficult to achieve fine classification of crops. Summary of the Invention

[0004] The embodiments of the present application provide a classification method, device, electronic device, storage medium and product, which can achieve refined classification of crops.

[0005] The technical solution of this application is achieved as follows:

[0006] In a first aspect, an embodiment of the present application provides a classification method, the method comprising:

[0007] Acquire first image data, second image data, and third image data corresponding to the target area within a preset time range;

[0008] Determining texture feature data of the target area corresponding to a first time unit based on the first image data, determining a plurality of spectral feature data and a plurality of exponential feature data of the target area corresponding to a second time unit based on the second image data, and determining a plurality of polarization feature data of the target area corresponding to a third time unit based on the third image data;

[0009] Determining a feature data set based on the texture feature data, the plurality of spectral feature data, the plurality of exponential feature data, and the plurality of polarization feature data;

[0010] The crop type of the target area is determined based on the feature data set and a general classification model.

[0011] In a second aspect, an embodiment of the present application provides a classification device, the device comprising:

[0012] An acquisition unit, configured to acquire first image data, second image data, and third image data corresponding to a target area within a preset time range;

[0013] A determination unit is configured to determine texture feature data of the target area corresponding to a first time unit based on the first image data, determine multiple spectral feature data and multiple exponential feature data of the target area corresponding to a second time unit based on the second image data, and determine multiple polarization feature data of the target area corresponding to a third time unit based on the third image data; determine a feature data set based on the texture feature data, the multiple spectral feature data, the multiple exponential feature data, and the multiple polarization feature data; and determine the crop type of the target area based on the feature data set and a general classification model.

[0014] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor and a memory; wherein:

[0015] The memory is used to store computer programs;

[0016] The processor is configured to call and run the computer program stored in the memory to execute the classification method described above.

[0017] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by at least one processor, the classification method as described above is implemented.

[0018] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements the classification method described above.

[0019] The embodiments of the present application provide a classification method, apparatus, electronic device, storage medium, and product. The method first obtains first image data, second image data, and third image data corresponding to a target area within a preset time range. Texture feature data corresponding to a first time unit of the target area is then determined based on the first image data. Multiple spectral feature data and multiple exponential feature data corresponding to a second time unit of the target area are determined based on the second image data. Polarization feature data corresponding to a third time unit of the target area are determined based on the third image data. A feature data set is determined based on the texture feature data, multiple spectral feature data, multiple exponential feature data, and multiple polarization feature data. Finally, the crop type of the target area is determined based on the feature data set and a general classification model. In this manner, corresponding feature calculations are performed on multiple different image data according to different time units. The obtained feature data is filtered and input into the classification model to further determine the crop type of the target area, thereby achieving refined classification of the crops. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings herein are incorporated into and constitute a part of this specification. These drawings illustrate embodiments consistent with the present application and, together with the specification, serve to illustrate the technical solutions of the present application. Obviously, the drawings described below are merely some embodiments of the present application. Those skilled in the art can, without inventive effort, derive other drawings from these drawings.

[0021] Figure 1 The figure is a flowchart of the rice recognition method based on the fusion of optical image and SAR time series data;

[0022] Figure 2 A schematic diagram of a classification method proposed in this embodiment of the application Figure 1 ;

[0023] Figure 3 Schematic diagram of the structure of the general classification model proposed in this application embodiment Figure 1 ;

[0024] Figure 4 Schematic diagram of the structure of the general classification model proposed in this application embodiment Figure 2 ;

[0025] Figure 5 A schematic diagram of a classification method proposed in this embodiment of the application Figure 2 ;

[0026] Figure 6 A schematic diagram of a classification method proposed in this embodiment of the application Figure 3 ;

[0027] Figure 7A schematic diagram of a classification method proposed in this embodiment of the application Figure 4 ;

[0028] Figure 8 This is a schematic diagram of the smoothed features proposed in the embodiment of the present application. Figure 1 ;

[0029] Figure 9 This is a schematic diagram of the smoothed features proposed in the embodiment of the present application. Figure 2 ;

[0030] Figure 10 This is a schematic diagram of the smoothed features proposed in the embodiment of the present application. Figure 3 ;

[0031] Figure 11 This is a schematic diagram of the smoothed features proposed in the embodiment of the present application. Figure 4 ;

[0032] Figure 12 This is a schematic diagram of the smoothed features proposed in the embodiment of the present application. Figure 5 ;

[0033] Figure 13 This is a schematic diagram of the smoothed features proposed in the embodiment of the present application. Figure 6 ;

[0034] Figure 14 This is a schematic diagram of the method for constructing quantile statistical features proposed in an embodiment of the present application;

[0035] Figure 15 This is a schematic diagram of the farmland plot proposed in the embodiment of this application;

[0036] Figure 16 This is a schematic diagram of a farmland plot obtained by the image segmentation method proposed in the embodiment of the present application;

[0037] Figure 17 This is a schematic diagram of a farmland plot obtained by the proximity window method proposed in the embodiment of this application;

[0038] Figure 18 A schematic diagram of the structure of a classification device proposed in an embodiment of the present application;

[0039] Figure 19 A schematic diagram of the structure of an electronic device proposed in an embodiment of the present application. DETAILED DESCRIPTION

[0040] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. It should be understood that the specific embodiments described herein are only used to explain the related applications and are not intended to limit the applications. It should also be noted that for ease of description, only the portions relevant to the related applications are shown in the drawings.

[0041] Crop distribution information is crucial for agricultural production management, economic development planning, and ecological protection policymaking. Remote sensing technology, with its advantages of wide monitoring coverage, short return cycles, rapid information acquisition, and low cost, can play a vital role in agricultural applications. However, traditional crop remote sensing classification studies focus on plains, and regional products primarily operate at resolutions of 10 meters or coarser, making them ineffective in meeting the demand for detailed crop distribution information in mountainous areas. Traditional remote sensing monitoring methods pose significant challenges in accurately mapping crop types in smallholder farming systems in mountainous areas, making them difficult to support the trend toward more refined and intelligent agricultural management.

[0042] Typically, when using optical remote sensing imagery for crop classification, the classification features used can be roughly divided into the following two categories: one utilizes the spectral characteristics of the crop itself for classification, such as the Normalized Difference Rapeseed Index. The other focuses on changes in planting or growth characteristics throughout a crop's growth cycle, constructing time series data of spectral or vegetation indices to extract phenological characteristics at different stages, such as sowing, growth, and maturity. However, optical remote sensing imagery data is often susceptible to weather conditions, and interpretation methods based on continuous optical images face significant challenges in their application. Compared to optical imagery, synthetic aperture radar signals can penetrate clouds and obtain temporally continuous data regardless of weather conditions, which, to a certain extent, allows for the extraction of crop phenological characteristics. However, the inevitable presence of noise in radar imagery negatively impacts crop identification, thereby limiting classification accuracy.

[0043] like Figure 1The following is a flow chart of a rice identification method based on the fusion of optical imagery and SAR (synthetic aperture radar) time-series data. This method primarily utilizes the spectral characteristics of single-temporal optical imagery and field survey data to coarsely classify the surveyed area, dividing different locations within the surveyed area into rice, water bodies, non-rice vegetation, and other surface areas. Based on the coarsely classified areas as rice, water bodies, and non-rice vegetation, multi-scale image segmentation is then performed using Bands 3, 4, …, and 8 of the single-view optical imagery and the Normalized Difference Vegetation Index (NDVI). Furthermore, the mean of plot feature values ​​is calculated based on plot boundaries, including several polarization features and / or texture features extracted from the time-series SAR imagery, to generate object-oriented time-series feature SAR data. Finally, machine learning algorithms such as CART (classification and regression tree) and SVM (support vector machine) are used to identify and classify rice within the surveyed area, determining the rice distribution (location, shape, and area of ​​rice plantings) within the surveyed area. Here, after feature sensitivity analysis and feature selection, rice business extraction can be carried out, and the key growth period characteristics of rice can be randomly extracted. The identification and classification of rice in the tested area can be completed through machine learning (CART, SVM), and the rice distribution in the tested area can be obtained.

[0044] However, the rice recognition method based on the fusion of optical imagery and SAR time series data has the following problems:

[0045] 1) When using the method of coarse classification and segmentation on single-period optical images for large-scale crop identification in mountainous areas, the accuracy of the coarse classification results and the precision of the segmented geographic objects will be greatly affected due to the spectral differences exhibited by rice at different growth stages. At the same time, selecting images based on the characteristics of the key growth period of rice will have a significant impact on the identification accuracy of other crops of the same period.

[0046] 2) Due to the influence of location and topography, the phenology of the same crop (such as rice) in a region often varies in time, sometimes by more than a month between the top and bottom of the mountain. This leads to inconsistent key growth periods and inconsistent features in the images at the same time. However, different crops (such as rice and corn) have similar characteristics during specific growth periods. Therefore, directly calculating the key growth period characteristics will lead to incorrect judgments.

[0047] 3) Obtaining key growth period characteristics through random sampling has great uncertainty and has a great impact on the accuracy of the results.

[0048] 4) Due to differences in the SAR mechanism, the accuracy of crop identification is lower than that of optical images of the same resolution. Insufficient use of optical images will make it difficult to improve the accuracy of easily confused crops.

[0049] To address this issue, this application proposes a classification method that determines texture feature data of a target area corresponding to a first time unit based on first acquired image data, multiple spectral feature data and multiple exponential feature data of the target area corresponding to a second time unit based on second acquired image data, and multiple polarization feature data of the target area corresponding to a third time unit based on third acquired image data. The method then determines the crop type in the target area based on the texture feature data, multiple spectral feature data, multiple exponential feature data, multiple polarization feature data, and a general classification model. In this way, the different image data are converted into corresponding feature data, and combined with the classification model to achieve refined classification of crops in mountainous areas.

[0050] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application.

[0051] The embodiments of the present application provide a classification method. It should be noted that the classification method proposed in the embodiments of the present application can be applied to a classification device or to an electronic device. This application does not specifically limit this. In the embodiments of the present application, the classification method proposed in the embodiments of the present application is exemplified by taking an electronic device as an example. The electronic device can be a variety of types of devices. For example, the electronic device can be, but is not limited to, any form of device such as a laptop computer.

[0052] Furthermore, Figure 2 A schematic diagram of a classification method proposed in this embodiment of the application Figure 1 ,like Figure 2 As shown, the classification method specifically includes the following steps 101 to 104:

[0053] Step 101: Acquire first image data, second image data, and third image data corresponding to a target area within a preset time range.

[0054] In an embodiment of the present application, the electronic device may obtain first image data, second image data, and third image data corresponding to the target area within a preset time range.

[0055] It should be noted that the preset time range can represent any time range. For example, the preset time range can represent one year, one month, or 10 days.

[0056] It should be noted that the first image data, the second image data, and the third image data can be acquired by any number of monitors with different resolutions. For example, the first image data can be acquired by a high-resolution remote sensing monitor; the second image data can be acquired by multiple medium-resolution remote sensing monitors; and the third image data can be acquired by a medium-resolution radar monitor.

[0057] It should be noted that the first image data may represent all image data collected by a high-resolution remote sensing monitor on the target area within one year. For example, the first image data may represent one period of image data collected by a high-resolution remote sensing monitor on the target area within one year.

[0058] It should be noted that the second image data may represent all image data collected by multiple medium-resolution remote sensing monitors on the target area every month. For example, the second image data may represent 12 periods of image data collected by multiple medium-resolution remote sensing monitors on the target area in one year.

[0059] It should be noted that the third image data may represent all image data collected by a medium-resolution radar monitor of the target area every 10 days. For example, the third image data may represent 36 periods of image data collected by a medium-resolution radar monitor of the target area within a year.

[0060] In the embodiments of the present application, the first image data, the second image data, and the third image data are all pre-processed image data, wherein the pre-processing includes at least one or more of the following data processing: geometric correction processing, radiometric correction processing, cloud detection and removal processing, position matching processing, terrain flattening processing, and denoising processing.

[0061] Furthermore, in an embodiment of the present application, when determining the second image data, the original image data corresponding to the target area is obtained; wherein the original image data includes the image data to be coordinated and the reference image data; based on the image data to be coordinated and the reference image data, the coordination parameters are determined; based on the coordination parameters, the image data to be coordinated is coordinated and processed to obtain the coordinated image data corresponding to the target area; and based on the coordinated image data and the reference image data, the second image data is determined.

[0062] Step 102: Determine texture feature data of the target area corresponding to a first time unit based on the first image data, determine multiple spectral feature data and multiple exponential feature data of the target area corresponding to a second time unit based on the second image data, and determine multiple polarization feature data of the target area corresponding to a third time unit based on the third image data.

[0063] In an embodiment of the present application, after obtaining the first image data, second image data, and third image data corresponding to the target area within a preset time range, the electronic device can further determine the texture feature data of the target area corresponding to the first time unit based on the first image data, determine the multiple spectral feature data and multiple exponential feature data of the target area corresponding to the second time unit based on the second image data, and determine the multiple polarization feature data of the target area corresponding to the third time unit based on the third image data.

[0064] It should be noted that the three time units can be different.

[0065] For example, the first time unit may represent 1 year as a time unit; the second time unit may represent 1 month as a time unit; and the third time unit may represent 10 days as a time unit.

[0066] It should be noted that the first, second, and third image data are all image data collected within the same preset time range. For example, the first image data is all image data collected over a year, representing one period of image data; the second image data is all image data collected over 12 months, representing 12 periods of image data; and the third image data is all image data collected over a year in 10-day increments, representing 36 periods of image data.

[0067] It should be noted that texture feature data may include multiple feature data such as entropy feature data and grayscale feature data.

[0068] It should be noted that the spectral characteristic data may include multiple characteristic data such as red spectrum characteristic data, green spectrum characteristic data, blue spectrum characteristic data and near-infrared spectrum characteristic data.

[0069] It should be noted that the index characteristic data may include normalized vegetation index characteristic data, enhanced vegetation index characteristic data, green chlorophyll vegetation index characteristic data, normalized difference vegetation aging index characteristic data, surface water index characteristic data, modified normalized difference water index characteristic data, normalized difference yellow index characteristic data, normalized difference soil index characteristic data, normalized difference tillage index characteristic data, water stress index characteristic data and other characteristic data.

[0070] It should be noted that the polarization characteristic data may include multiple characteristic data such as vertical polarization characteristic data and cross-polarization characteristic data.

[0071] It should be noted that, in the embodiment of the present application, the execution entity may use a Savitzky-Golay filter to perform noise smoothing on multiple polarization feature data.

[0072] Step 103: Determine a feature data set based on the texture feature data, the plurality of spectral feature data, the plurality of exponential feature data, and the plurality of polarization feature data.

[0073] In an embodiment of the present application, after determining the texture feature data of the target area corresponding to the first time unit based on the first image data, determining the multiple spectral feature data and the multiple exponential feature data of the target area corresponding to the second time unit based on the second image data, and determining the multiple polarization feature data of the target area corresponding to the third time unit based on the third image data, the electronic device can further determine the feature data set based on the texture feature data, the multiple spectral feature data, the multiple exponential feature data and the multiple polarization feature data.

[0074] Furthermore, in an embodiment of the present application, multiple spectral feature data are classified, statistical values ​​of each type of spectral feature data are obtained, and a first feature data set is determined based on the statistical values ​​of each type of spectral feature data; multiple exponential feature data are classified, statistical values ​​of each type of exponential feature data are obtained, and a second feature data set is determined based on the statistical values ​​of each type of exponential feature data; multiple polarization feature data are classified, statistical values ​​of each type of polarization feature data are obtained, and a third feature data set is determined based on the statistical values ​​of each type of polarization feature data; and a feature data set is determined based on texture feature data, the first feature data set, the second feature data set, and the third feature data set.

[0075] It should be noted that the statistical value of each type of spectral feature data can represent any numerical value. For example, the statistical value can represent various numerical values ​​such as the minimum value, maximum value, average value, standard deviation, skewness, kurtosis, and percentile value of each type of spectral feature data.

[0076] It should be noted that the statistical value of each type of index feature data can represent any form of numerical value. For example, the statistical value can represent various numerical values ​​such as the minimum value, maximum value, average value, standard deviation, skewness, kurtosis, and percentile value of each type of index feature data.

[0077] It should be noted that the statistical value of each type of polarization characteristic data can represent any numerical value. For example, the statistical value can represent various numerical values ​​such as the minimum value, maximum value, average value, standard deviation, skewness, kurtosis, and percentile value of each type of polarization characteristic data.

[0078] Furthermore, in an embodiment of the present application, based on texture feature data, a first feature data set, a second feature data set, and a third feature data set, feature data to be selected is determined; the feature data to be selected is screened to determine a plurality of feature data to be clustered; the plurality of feature data to be clustered is clustered to obtain a plurality of cluster feature data; and a feature data set is determined based on the feature data with the highest index in each cluster feature data.

[0079] It should be noted that the plurality of feature data to be clustered may represent feature data with a high MDI (Medium Dependent Interface).

[0080] It should be noted that the plurality of cluster feature data are feature data determined based on a rank correlation algorithm and a maximum depth threshold.

[0081] It should be noted that the feature data set may represent a data set consisting of feature data with the highest MDI in each cluster.

[0082] Step 104: Determine the crop type of the target area based on the feature data set and the general classification model.

[0083] In an embodiment of the present application, after determining a feature data set based on texture feature data, multiple spectral feature data, multiple exponential feature data, and multiple polarization feature data, the electronic device can further determine the crop type of the target area based on the feature data set and a general classification model.

[0084] It should be noted that the general classification model can be a deep metric space classification model. This general classification model is a convolutional neural network framework based on the attention mechanism and deep metric learning. While introducing spatial neighborhood information, it seeks a more effective Euclidean space. In the new feature space, the crops are screened and classified by having the minimum intra-class distance and the maximum intra-class distance. Among them, the deep metric space classification model consists of 3 convolution blocks, 1 attention mechanism module, 1 global average pooling layer and 1 fully connected layer; Among them, 3 convolution blocks and 1 attention mechanism module constitute the metric space, such as Figure 3 The structure of the general classification model shown Figure 1 , the model embeds crop samples into a specific metric space, in which the distance between any two samples can be represented. A two-dimensional convolutional neural network structure is used as the embedded network of the general classification model. In order to comprehensively utilize spectral features and spatial information, this application uses each crop sample point pixel and its adjacent pixels within a P×P window as the sample input of the network. Figure 4 The structure of the general classification model shown Figure 2In the embedding module of this model, there are three CB convolution blocks and one CBAM attention mechanism module, which are mainly used for mapping and selecting a better depth metric space. The CB module implements the mapping of the depth metric space, which contains D 1×1 convolution kernels, a batch normalization layer and a ReLU (Rectified Linear Unit) nonlinear activation function.

[0085] Furthermore, in an embodiment of the present application, the electronic device inputs the crop sample data and feature data set into a general classification model, outputting at least one Euclidean distance corresponding to each sub-region within the target region; wherein each sub-region corresponds to at least one crop type, and each crop type corresponds to one Euclidean distance. Based on the at least one Euclidean distance, the crop type corresponding to each sub-region is determined.

[0086] Furthermore, in an embodiment of the present application, for any sub-area, the electronic device determines a first ratio value based on the first Euclidean distance and the second Euclidean distance corresponding to the sub-area; wherein the first Euclidean distance and the second Euclidean distance are the two largest Euclidean distances corresponding to the sub-area, and the first Euclidean distance is greater than the second Euclidean distance; when the first ratio value is greater than the first threshold, the electronic device determines the crop type corresponding to the first Euclidean distance as the crop type of the sub-area.

[0087] Furthermore, in an embodiment of the present application, when the first ratio value is less than or equal to the first threshold value, the electronic device determines the second threshold value based on the area parameter of the sub-region; when the first ratio value is greater than the second threshold value, the electronic device determines the crop type corresponding to the first Euclidean distance as the crop type of the sub-region; when the first ratio value is less than or equal to the second threshold value, the electronic device determines the crop type corresponding to the second Euclidean distance as the crop type of the sub-region.

[0088] For example, assume that a sub-region of the target region contains wheat and rice. The wheat corresponds to the first Euclidean distance, and the rice corresponds to the second Euclidean distance. The first Euclidean distance is greater than the second Euclidean distance. If a first ratio of the first Euclidean distance to the second Euclidean distance is greater than a first threshold, the crop type of the sub-region is determined to be wheat.

[0089] Excluding the sub-region containing wheat, if the first ratio is less than or equal to the first threshold, a second threshold for crops is set based on the area of ​​the remaining region. Assuming that the remaining region contains oats and corn, oats are assigned the first Euclidean distance, while corn is assigned the second Euclidean distance. If the first ratio of the first to second Euclidean distances is greater than the second threshold, the crop type for this sub-region is oats; otherwise, it is corn.

[0090] An embodiment of the present application provides a classification method, in which an electronic device first obtains first image data, second image data, and third image data corresponding to a target area within a preset time range; then, based on the first image data, determines texture feature data of the target area corresponding to a first time unit; based on the second image data, determines multiple spectral feature data and multiple exponential feature data of the target area corresponding to a second time unit; and based on the third image data, determines multiple polarization feature data of the target area corresponding to a third time unit; determines a feature data set based on the texture feature data, multiple spectral feature data, multiple exponential feature data, and multiple polarization feature data; and finally, determines the crop type of the target area based on the feature data set and a general classification model. In this way, corresponding feature calculations are performed on multiple different image data according to different time units, and the obtained feature data is filtered and input into the classification model to further determine the crop type of the target area, thereby achieving refined classification of the crops.

[0091] Based on the above embodiments, in some embodiments, Figure 5 A schematic diagram of a classification method proposed in this embodiment of the application Figure 2 ,like Figure 5 As shown, the implementation of obtaining the second image data may include steps 201 to 204:

[0092] Step 201: Acquire original image data corresponding to the target area; wherein the original image data includes image data to be coordinated and reference image data.

[0093] In an embodiment of the present application, the electronic device may acquire original image data corresponding to the target area; wherein the original image data includes image data to be coordinated and reference image data.

[0094] It should be noted that the image data to be coordinated may represent image data with changed spatial positions. For example, the image data to be coordinated refers to multiple image data collected by multiple medium-resolution remote sensing monitors at the same location in the target area, but with different spatial positions.

[0095] It should be noted that the reference image data may represent image data whose spatial position has not changed. For example, the reference image data refers to multiple image data collected by multiple medium-resolution remote sensing monitors at the same location in the target area, with the spatial position being exactly the same.

[0096] Step 202: Determine coordination parameters based on the image data to be coordinated and the reference image data.

[0097] In an embodiment of the present application, after acquiring original image data corresponding to the target area, wherein the original image data includes the image data to be coordinated and the reference image data, the electronic device may further determine coordination parameters based on the image data to be coordinated and the reference image data.

[0098] It should be noted that the coordination parameters are calculated based on the original pixel values ​​of the image data to be coordinated and the original pixel values ​​of the reference image data.

[0099] It should be noted that the coordination parameter can represent the parameters of the polynomial regression coordination function, wherein the polynomial regression coordination function can be defined according to the following formula:

[0100] y=ax 2 + bx+c (1)

[0101] In formula (1), a, b, and c are coordination parameters; x is the image data to be coordinated.

[0102] It should be noted that the polynomial regression coordination function is constructed as follows:

[0103] First, based on the reference image data, the target area is divided to determine multiple sub-target areas.

[0104] Then, obtain the point data corresponding to each sub-target area.

[0105] Finally, a coordination function is constructed based on the image data to be coordinated and the reference image data corresponding to each point data.

[0106] Step 203: Perform coordination processing on the image data to be coordinated based on the coordination parameters to obtain coordinated image data corresponding to the target area.

[0107] In an embodiment of the present application, after determining the coordination parameters based on the image data to be coordinated and the reference image data, the electronic device may further coordinate the image data to be coordinated based on the coordination parameters to obtain coordinated image data corresponding to the target area.

[0108] It should be noted that by inputting the image data to be coordinated into the coordination function, the coordinated image data corresponding to the target area can be output.

[0109] Step 204: Determine second image data based on the coordinated image data and the reference image data.

[0110] In an embodiment of the present application, after performing coordination processing on the image data to be coordinated based on the coordination parameters to obtain coordinated image data corresponding to the target area, the electronic device may further determine second image data based on the coordinated image data and the reference image data.

[0111] It should be noted that the second image data may represent a plurality of image data with the same spatial position collected by a plurality of medium-resolution remote sensing monitors at the same position of the target area.

[0112] In the embodiment of the present application, based on the coordination parameters determined by the image data to be coordinated and the reference image data, the image data to be coordinated is coordinated to obtain the second image data. In this way, by performing the harmonization processing on the multi-source medium-resolution remote sensing image data, a remote sensing image dataset with consistent spatial positions is formed.

[0113] Based on the above embodiments, in some embodiments, Figure 6 A schematic diagram of a classification method proposed in this embodiment of the application Figure 3 ,like Figure 6 As shown, the implementation of determining the crop type of the target area based on the feature data set and the general classification model may include the following steps:

[0114] Step 104a: Input the crop sample data and feature data set into a general classification model, and output at least one Euclidean distance corresponding to each sub-region in the target region; wherein each sub-region corresponds to at least one crop type, and one crop type corresponds to one Euclidean distance.

[0115] In an embodiment of the present application, after determining a feature data set based on texture feature data, multiple spectral feature data, multiple exponential feature data, and multiple polarization feature data, the electronic device can further input the crop sample data and the feature data set into a general classification model, and output at least one Euclidean distance corresponding to each sub-region in the target area; wherein each sub-region corresponds to at least one crop type, and one crop type corresponds to one Euclidean distance.

[0116] Step 104b: Determine the crop type corresponding to each sub-region based on at least one Euclidean distance.

[0117] In an embodiment of the present application, after inputting the crop sample data and feature data set into a general classification model and outputting at least one Euclidean distance corresponding to each sub-region in the target area, the electronic device can further determine the crop type corresponding to each sub-region based on the at least one Euclidean distance.

[0118] Furthermore, in an embodiment of the present application, for any sub-region, a first ratio value is determined based on the first Euclidean distance and the second Euclidean distance corresponding to the sub-region; wherein the first Euclidean distance and the second Euclidean distance are the two largest Euclidean distances corresponding to the sub-region, and the first Euclidean distance is greater than the second Euclidean distance; when the first ratio value is greater than the first threshold value, the crop type corresponding to the first Euclidean distance is determined as the crop type of the sub-region.

[0119] Furthermore, in an embodiment of the present application, when the first ratio value is less than or equal to the first threshold value, the second threshold value is determined based on the area parameter of the sub-region; when the first ratio value is greater than the second threshold value, the crop type corresponding to the first Euclidean distance is determined as the crop type of the sub-region; when the first ratio value is less than or equal to the second threshold value, the crop type corresponding to the second Euclidean distance is determined as the crop type of the sub-region.

[0120] It should be noted that, in practical applications, the second threshold R is an empirical value, for example, 1≤R≤2.

[0121] For example, assume that one sub-region of the target region contains wheat and rice. The wheat corresponds to the first Euclidean distance, and the rice corresponds to the second Euclidean distance. The first Euclidean distance is greater than the second Euclidean distance. If the first ratio of the first Euclidean distance to the second Euclidean distance is greater than 2, the crop type of the sub-region is determined to be wheat.

[0122] Excluding the subregion containing wheat, if the first ratio is less than or equal to 2, the second threshold for crops is set to 1 based on the area of ​​the remaining region. Assuming that the remaining region contains oats and corn, oats correspond to the first Euclidean distance, and corn corresponds to the second Euclidean distance. If the first ratio of the first to second Euclidean distances is greater than 1, the crop type of this subregion is oats; otherwise, it is corn.

[0123] In this embodiment, crop sample data and a feature data set are input into a general classification model. At least one Euclidean distance corresponding to each subregion within the target region is output, and the crop type corresponding to each subregion is determined based on the at least one Euclidean distance. In this manner, based on the general classification model, the first threshold, and the second threshold, the crop type of each subregion within the target region is accurately classified.

[0124] Based on the above embodiments, the classification method provided in the embodiments of the present application is described in detail below in combination with specific application scenarios. Figure 7 A schematic diagram of a classification method proposed in this embodiment of the application Figure 4 ,like Figure 7 As shown, the classification method specifically includes the following steps 301 to 306:

[0125] The obtained high-resolution optical annual synthetic remote sensing image dataset (first image data), multi-source medium-resolution time series optical remote sensing image dataset, and medium-resolution radar time series remote sensing image dataset (third image data) are all pre-processed datasets.

[0126] It should be noted that pre-processing includes geometric correction, radiometric correction, cloud detection and removal, position matching between multiple image periods, and terrain flattening and denoising of SAR data. The high-resolution image data is synthesized by optimizing and selecting cloud-free data from within one year.

[0127] Step 301: Multi-source data band coordination.

[0128] Due to the frequent cloudiness and rain in mountainous areas, it is difficult for a single satellite to achieve full coverage in both spatial and temporal dimensions. Therefore, this application collects a variety of satellite image data with a spatial resolution of 10 meters to 30 meters. For different types of satellite image data, it is necessary to coordinate the multi-source data bands. The specific coordination method is as follows:

[0129] First, clustering is performed based on the spectral characteristics of coordinated benchmark type satellite images (benchmark image data), and the analysis area is divided into multiple regions based on spectral similarity.

[0130] Secondly, random sampling is used to generate matching points in each area.

[0131] Then, based on the spectral feature values ​​of the same matching points on different types of images, the coordination function of different types of satellite images in different regions to the benchmark image is constructed using polynomial regression.

[0132] Finally, the coordination function is used to unify the multi-source medium-resolution optical remote sensing image data to form a fused fragmented medium-resolution optical remote sensing image dataset (second image data).

[0133] It should be noted that the coordination function of polynomial regression can be defined according to formula (1).

[0134] It should be noted that the polynomial regression coordination function is constructed as follows:

[0135] First, based on the reference image data, the target area is divided to determine multiple sub-target areas.

[0136] Then, obtain the point data corresponding to each sub-target area.

[0137] Finally, a coordination function is constructed based on the image data to be coordinated and the reference image data corresponding to each point data.

[0138] Step 302: Calculate key features.

[0139] Here, the key feature calculations include: image texture feature calculation, monthly synthetic spectrum and index feature calculation, and 10-day synthetic polarization feature calculation.

[0140] Under the constraints of farmland plot masks (farmland plot data), a high-resolution optical annual synthetic remote sensing image dataset is used to calculate the texture features within the spatial range of farmland plots. The texture features mainly include entropy and grayscale correlation.

[0141] Under the constraints of farmland plot masks (farmland plot data), a fused, fragmented, medium-resolution optical remote sensing image dataset was used to calculate monthly composite spectral and index features within the spatial range of the farmland plots. Spectral features include red, green, blue, and near-infrared spectral values, calculated using the within-plot mean method; index features include the remote sensing indices shown in Table 1.

[0142] Table 1. Remote sensing index calculation

[0143]

[0144] Enhanced vegetation index (EVI); Green chlorophyll vegetation index (GCVI); Normalized differential vegetation aging index (NDVAI); Land surface water index (LSWI); Modified normalized differential water index (MNDWI); Normalized differential yellow index (NDYI); Normalized differential soil index (NDSI); Normalized differential tillage index (NDTI); and Water stress index (WSI). For cloudy and rainy mountainous areas, it is difficult to construct time series with intervals of 5 or 10 days using optical remote sensing images. Based on the above feature calculations, the median composite method was used to calculate the index values ​​shown in Table 1 for all six effective optical observation bands (red, green, blue, near-infrared, shortwave infrared 1, and shortwave infrared 2). In Table 1, Nir refers to near infrared and Swir refers to shortwave infrared.

[0145] Using a medium-resolution radar time series remote sensing image dataset, we calculated the polarization characteristics of farmland plots. The calculation formula for the polarization characteristics is as follows:

[0146]

[0147] In formula (2), and are the backscattering coefficients of VH and VV in the logarithmic domain.

[0148] It should be noted that there are noise or abnormal mutation values ​​in the uniform seamless time series remote sensing image data synthesized by 10-day median value. Therefore, this application smoothes the time series of SAR features through SG filtering, that is, the SG filter with a 2×2 window is used to smooth the noise of the SAR time series generated by the linear moving interpolation method to eliminate noise or abnormal mutation values. Figure 8 Schematic diagram of the smoothed features shown Figure 1 ;like Figure 9 Schematic diagram of the smoothed features shown Figure 2 ;like Figure 10 Schematic diagram of the smoothed features shown Figure 3 ;like Figure 11 Schematic diagram of the smoothed features shown Figure 4 ;like Figure 12 Schematic diagram of the smoothed features shown Figure 5 ;like Figure 13 Schematic diagram of the smoothed features shown Figure 6 .

[0149] Step 303: Calculate the time statistics of the feature.

[0150] In order to capture all valid observations and characteristics of crop phenology throughout the growing season, temporal statistics were calculated for the monthly composite spectral and index characteristics and the 10-day composite polarization characteristics to describe the heterogeneity of different characteristics.

[0151] It should be noted that the time statistics include but are not limited to the minimum value, maximum value, mean value, standard deviation, skewness, kurtosis and nth percentile, where n is a positive integer.

[0152] like Figure 14 The diagram below shows how to construct quantile statistics. The left side shows the observed images of the time series, with each parallelogram representing one image period. The gray square area represents a pixel (with a spatial location), that is, the pixel at the same position in each image period. These values ​​are read, corresponding to the points in the two figures on the right. The first figure on the right shows a graph of all pixel values ​​read, with time as the horizontal axis and pixel value as the vertical axis. This graph shows how the observed values ​​change over time. The second figure on the right shows a graph of all pixel values ​​read, with percentiles as the horizontal axis and pixel value as the vertical axis. This graph shows the percentile position of the observed value.

[0153] Step 304: Feature optimization.

[0154] Initial screening: feature selection based on MDI importance.

[0155] The importance of all data features was evaluated by MDI, which was calculated by the random forest classifier. The top 80 data features were obtained based on MDI sorting and screening.

[0156] Re-screening: The rank correlation algorithm is used to perform hierarchical clustering on the first 80 features to reduce the multicollinearity that may exist in the initially selected features. The first 80 features of the initial screening are grouped into 32 clusters through the maximum depth threshold (which can be set to 1).

[0157] Finally, the data feature with the highest MDI in each cluster is retained.

[0158] Through this hierarchical clustering screening method, the multicollinearity between the selected crop classification features is significantly reduced, and the redundancy of the input information is greatly reduced.

[0159] Step 305: Deep metric learning.

[0160] A deep metric learning model (general classification model) is constructed for samples of complex mountainous crops to identify multiple types of crops.

[0161] like Figure 3 As shown, the model embeds complex mountain crop samples into a specific metric space, in which the distance between any two samples can be represented. A two-dimensional convolutional neural network structure is used as the embedded network for the deep metric learning model. To comprehensively utilize spectral features and spatial information, this application uses each crop sample pixel and its neighboring pixels within a P×P window as sample inputs to the network. Therefore, the size of each sample is N×P×P, where N is the number of features retained in the feature construction and optimization experiments, a total of 32.

[0162] like Figure 4 As shown in Figure 3, the embedding module of this model contains three CB convolution blocks and one CBAM attention mechanism module, which are mainly used for mapping and selecting a better depth metric space. The CB module implements the mapping of the depth metric space and contains D 1×1 convolution kernels, a batch normalization layer, and a ReLU nonlinear activation function.

[0163] Because our model uses a 1×1 convolution kernel, the number of network parameters is significantly reduced, which is suitable for the limited crop sample size in mountainous areas. The input of the neighborhood window inevitably introduces interfering pixels such as houses and roads. To suppress these interfering pixels, a CBAM (Convolutional Block Attention Module) attention mechanism is introduced into the model.

[0164] It should be noted that in order to embed the metric features learned by the model into the Euclidean space, a global average pooling layer and a fully connected layer are added to the last part of the model to obtain the final mapping space under the fully connected layer.

[0165] Step 306: Multi-type layered fusion.

[0166] Based on the deep metric learning model, the possible type of each plot can be judged according to the size of the Euclidean distance of the sample in the metric space.

[0167] It should be noted that for interference from similar crops, such as wheat and oats, some plots have similar probabilities for multiple types. Therefore, this application uses regional multi-crop planting area statistics (such as the annual planting area of ​​rice, corn, and other crops in counties and districts in local statistical yearbooks) to constrain the mapping of multi-type crop distribution maps.

[0168] In order to solve the problem of interference from similar crops, this application proposes the following solution:

[0169] First, the Euclidean distance values ​​of multiple types of plots are output by the deep metric learning model, and the two types with the largest Euclidean distance values ​​on each plot are obtained as suspected types.

[0170] Next, calculate the Euclidean distance ratio of the suspected type:

[0171] R= D max / D min (3)

[0172] In formula (3), D max is the maximum Euclidean distance value, D min is the minimum Euclidean distance value.

[0173] Again, the plots with R greater than 2 are screened out, and the type corresponding to the large Euclidean distance is used as the crop type of this plot.

[0174] Then, the statistical area of ​​the identified crop type is subtracted from the statistical area data of different crops in the region to obtain the remaining crop area.

[0175] Finally, the plots between 1≤R≤2 are divided into different levels according to the size of the R value and the mapping R value threshold of different crops is set. The area with R≥threshold is set as D max The corresponding type, otherwise set to D min Corresponding types. Further obtain the distribution map of multiple types of crops.

[0176] It should be noted that, in addition, when there is only a single data source, the band coordination of the multi-source optical data in this application can be supplemented by interpolating the adjacent time data. If the farmland mask data (farmland plot data) in this application is missing, the farmland mask data can be replaced by image segmentation or adjacent window as the calculation range, such as Figure 15 Schematic diagram of the farmland plot shown; Figure 16 The schematic diagram of farmland plots obtained by image segmentation method is shown; Figure 17 The diagram of the farmland plot obtained using the proximity window method is shown. In the feature optimization step of this application, the number of features to be screened can be selected from other values. The statistical data on planted area in this application can be replaced by data estimated statistically after a sampling survey; the setting of the R value can be adjusted based on the statistical information of the D value of the entire region.

[0177] The classification method of the present application can achieve the following effects: by obtaining multi-source image data, the impact of clouds and rain on observation data can be reduced. By smoothing the time series features, the variation caused by data anomalies can be reduced. By using metric learning to perform linear changes in the feature space, the problem of identifying differences between multiple categories can be solved. The present application utilizes three types of image data. In addition to the time series SAR data of the crop growing year, it also utilizes incomplete time series optical data of the crop growing year (due to spatial and temporal loss caused by clouds and rain) and high spatial resolution image data fused throughout the crop growing year. Compared with the farmland boundary data obtained by image segmentation methods, the real planting boundaries obtained by the present application through investigation have a more reliable data source. Compared with the identification of a single crop, the present application can simultaneously and accurately identify multiple crops planted in the current year. In addition to image features, the features used in the model of the present application also include feature time statistical features calculated based on image features. The present application is based on a convolutional neural network framework based on the attention mechanism and deep metric learning. While introducing spatial neighborhood information, it seeks a more effective Euclidean space. In the new feature space, feature screening and classification are performed based on the minimum intra-class distance and maximum intra-class distance of crops. This application combines regional statistical information to define crop category probability thresholds, and then forms a planting structure distribution product.

[0178] Based on the same inventive concept as the above embodiment, another embodiment of the present application provides a classification device, Figure 18 A schematic diagram of the structure of a classification device is shown in FIG. Figure 18 As shown, the classification device 300 includes: an acquisition unit 3001 and a determination unit 3002.

[0179] An acquisition unit 3001 is configured to acquire first image data, second image data, and third image data corresponding to a target area within a preset time range;

[0180] The determination unit 3002 is used to determine the texture feature data of the target area corresponding to the first time unit based on the first image data, determine multiple spectral feature data and multiple exponential feature data of the target area corresponding to the second time unit based on the second image data, and determine multiple polarization feature data of the target area corresponding to the third time unit based on the third image data; determine a feature data set based on the texture feature data, the multiple spectral feature data, the multiple exponential feature data, and the multiple polarization feature data; and determine the crop type of the target area based on the feature data set and a general classification model.

[0181] The present application also provides a schematic diagram of the structure of an electronic device, such as Figure 19 As shown, the electronic device 40 proposed in the embodiment of the present application may include a processor 41 and a memory 42 configured to store a computer program that can be run on the processor. Furthermore, the electronic device 40 may also include a communication interface 43 and a bus 44 for connecting the processor 41, the memory 42 and the communication interface 43.

[0182] In the embodiment of the present application, the processor 41 may be at least one of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, and a microprocessor. It is understood that for different devices, the electronic device used to implement the above-mentioned processor function may also be other, and the embodiment of the present application does not specifically limit this.

[0183] In the embodiment of the present application, the memory 42 is used to store instructions and data. The memory 42 can be connected to the processor 41, wherein the memory 42 is used to store executable program code, which includes computer operation instructions.

[0184] In the embodiment of the present application, the bus 44 is used to connect the communication interface 43, the processor 41 and the memory 42, as well as the mutual communication between these devices.

[0185] Furthermore, in an embodiment of the present application, the above-mentioned processor 41 is used to obtain first image data, second image data, and third image data corresponding to the target area within a preset time range; determine the texture feature data of the target area corresponding to the first time unit based on the first image data, determine multiple spectral feature data and multiple index feature data of the target area corresponding to the second time unit based on the second image data, and determine multiple polarization feature data of the target area corresponding to the third time unit based on the third image data; determine a feature data set based on the texture feature data, the multiple spectral feature data, the multiple index feature data, and the multiple polarization feature data; and determine the crop type of the target area based on the feature data set and a general classification model.

[0186] In practical applications, the memory 42 may be a volatile memory, such as a random-access memory (RAM); or a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor 41.

[0187] The embodiments of the present application provide a classification device and an electronic device, which obtain first image data, second image data, and third image data corresponding to a target area within a preset time range; determine texture feature data of the target area corresponding to a first time unit based on the first image data; determine multiple spectral feature data and multiple exponential feature data of the target area corresponding to a second time unit based on the second image data; and determine multiple polarization feature data of the target area corresponding to a third time unit based on the third image data; determine a feature data set based on the texture feature data, the multiple spectral feature data, the multiple exponential feature data, and the multiple polarization feature data; and determine the crop type of the target area based on the feature data set and a general classification model. In this way, corresponding feature calculations are performed on multiple different image data according to different time units, and the obtained feature data are filtered and input into the classification model to further determine the crop type of the target area, thereby achieving refined classification of crops.

[0188] An embodiment of the present application further provides a computer-readable storage medium on which a program is stored, and when the program is executed by a processor, the classification method described above is implemented.

[0189] Specifically, the program instructions corresponding to a classification method in this embodiment can be stored on a storage medium such as a CD, a hard disk, or a USB flash drive. When the program instructions corresponding to a classification method in the storage medium are read or executed by an electronic device, the following steps are included:

[0190] Acquire first image data, second image data, and third image data corresponding to the target area within a preset time range;

[0191] Determining texture feature data of the target area corresponding to a first time unit based on the first image data, determining a plurality of spectral feature data and a plurality of exponential feature data of the target area corresponding to a second time unit based on the second image data, and determining a plurality of polarization feature data of the target area corresponding to a third time unit based on the third image data;

[0192] Determining a feature data set based on the texture feature data, the plurality of spectral feature data, the plurality of exponential feature data, and the plurality of polarization feature data;

[0193] The crop type of the target area is determined based on the feature data set and a general classification model.

[0194] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0195] The present application is described with reference to the implementation flow charts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flow charts and / or block diagrams, as well as the combination of processes and / or boxes in the flow charts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the implementation flow charts. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0196] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which is implemented in the implementation flow diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0197] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process described in the flowchart. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0198] The above description is merely a preferred embodiment of the present application and is not intended to limit the scope of protection of the present application.

Claims

1. A classification method, characterized in that The method comprises: Acquire first image data, second image data, and third image data corresponding to a target area within a preset time range; the first image data represents all image data of the target area collected by a high-resolution remote sensing monitor; the second image data represents all image data of the target area collected by multiple medium-resolution remote sensing monitors; and the third image data represents all image data of the target area collected by a medium-resolution radar monitor; Determining texture feature data of the target area corresponding to a first time unit based on the first image data, determining a plurality of spectral feature data and a plurality of exponential feature data of the target area corresponding to a second time unit based on the second image data, and determining a plurality of polarization feature data of the target area corresponding to a third time unit based on the third image data; Determining a feature data set based on the texture feature data, the plurality of spectral feature data, the plurality of exponential feature data, and the plurality of polarization feature data; Inputting crop sample data and the feature data set into the feature data set, outputting at least one Euclidean distance corresponding to each sub-region in the target region; wherein each sub-region corresponds to at least one crop type, and one crop type corresponds to one Euclidean distance; for any sub-region, determining a first ratio value based on a first Euclidean distance and a second Euclidean distance corresponding to the sub-region; wherein the first Euclidean distance and the second Euclidean distance are the two largest Euclidean distances corresponding to the sub-region, and the first Euclidean distance is greater than the second Euclidean distance; when the first ratio value is greater than a first threshold, determining the crop type corresponding to the first Euclidean distance as the crop type of the sub-region; When the first ratio value is less than or equal to the first threshold, the second threshold is determined according to the area parameter of the sub-region; when the first ratio value is greater than the second threshold, the crop type corresponding to the first Euclidean distance is determined as the crop type of the sub-region; when the first ratio value is less than or equal to the second threshold, the crop type corresponding to the second Euclidean distance is determined as the crop type of the sub-region.

2. The method according to claim 1, characterized in that Acquiring second image data, including: Acquire original image data corresponding to the target area; wherein the original image data includes image data to be coordinated and reference image data; determining coordination parameters based on the image data to be coordinated and the reference image data; Performing coordination processing on the image data to be coordinated based on the coordination parameters to obtain coordinated image data corresponding to the target area; The second image data is determined based on the coordinated image data and the reference image data.

3. The method according to claim 1 or 2, characterized in that The determining of a feature data set based on the texture feature data, the plurality of spectral feature data, the plurality of exponential feature data, and the plurality of polarization feature data comprises: Classifying the plurality of spectral feature data, obtaining statistical values ​​of each category of spectral feature data, and determining a first feature data set according to the statistical values ​​of each category of spectral feature data; Classifying the plurality of index feature data, obtaining statistical values ​​of each category of index feature data, and determining a second feature data set based on the statistical values ​​of each category of index feature data; Classifying the plurality of polarization feature data, obtaining statistical values ​​of each type of polarization feature data, and determining a third feature data set based on the statistical values ​​of each type of polarization feature data; A feature data set is determined based on the texture feature data, the first feature data set, the second feature data set, and the third feature data set.

4. The method according to claim 3, characterized in that The determining of a feature data set based on the texture feature data, the first feature data set, the second feature data set, and the third feature data set includes: Determining feature data to be selected based on the texture feature data, the first feature data set, the second feature data set, and the third feature data set; Screening the feature data to be selected to determine a plurality of feature data to be clustered; Performing clustering processing on the plurality of feature data to be clustered to obtain a plurality of cluster feature data; The feature data set is determined based on the feature data with the highest index in each cluster feature data.

5. A classification device, characterized in that: The device comprises: an acquisition unit, configured to acquire first image data, second image data, and third image data corresponding to a target area within a preset time range; the first image data representing all image data of the target area collected by a high-resolution remote sensing monitor; the second image data representing all image data of the target area collected by multiple medium-resolution remote sensing monitors; and the third image data representing all image data of the target area collected by a medium-resolution radar monitor; A determination unit is configured to determine texture feature data of the target area corresponding to a first time unit based on the first image data, determine multiple spectral feature data and multiple exponential feature data of the target area corresponding to a second time unit based on the second image data, and determine multiple polarization feature data of the target area corresponding to a third time unit based on the third image data; determine a feature data set based on the texture feature data, the multiple spectral feature data, the multiple exponential feature data, and the multiple polarization feature data; input crop sample data and the feature data set into the feature data set, and output at least one Euclidean distance corresponding to each sub-area in the target area; wherein each sub-area corresponds to at least one crop type, and one crop type corresponds to one Euclidean distance; for any of the sub-areas, according to The first Euclidean distance and the second Euclidean distance corresponding to the sub-region determine a first ratio value; wherein, the first Euclidean distance and the second Euclidean distance are the two largest Euclidean distances corresponding to the sub-region, and the first Euclidean distance is greater than the second Euclidean distance; when the first ratio value is greater than a first threshold, the crop type corresponding to the first Euclidean distance is determined as the crop type of the sub-region; when the first ratio value is less than or equal to the first threshold, the second threshold is determined according to the area parameter of the sub-region; when the first ratio value is greater than the second threshold, the crop type corresponding to the first Euclidean distance is determined as the crop type of the sub-region; when the first ratio value is less than or equal to the second threshold, the crop type corresponding to the second Euclidean distance is determined as the crop type of the sub-region.

6. An electronic device, characterized in that: The electronic device includes: a processor and a memory, the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to perform the classification method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by at least one processor, the classification method according to any one of claims 1 to 4 is implemented.

8. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the classification method according to any one of claims 1 to 4 is implemented.

Images