Polarimetric SAR crop classification method, system, equipment and medium based on multi-feature joint time series matching

Abstract

The present invention discloses a polarimetric SAR crop classification method, system, device and medium based on multi-feature joint time series matching, the method comprises the following steps: obtaining a multi-phase polarimetric SAR data set of a crop region, extracting multiple features to obtain each feature time series, and calculating each polarimetric feature discrimination degree; weighting the Euclidean distance with each feature discrimination degree, and using it as the base distance for aligning the feature time series curve to perform path matching on the feature time series of the crop to be classified; dividing the feature time series of the crop to be classified into subsequences according to the phenological calendar, and calculating the matching distance similarity between the crop to be classified and each standard crop based on the subsequence similarity, the subsequence mean time series feature and each feature discrimination degree, and determining the standard crop category corresponding to the minimum value as the category of the crop to be classified. The present invention makes full use of the time-varying characteristics and phenological information of crops, can solve the problem of phase shift caused by phenological uncertainty, and improve the discrimination degree of crop categories.

Description

Technical Field

[0001] The present invention relates to crop classification, and in particular to a polarimetric SAR crop classification method, system, equipment and medium based on multi-feature joint time series matching taking into account phenological shift. Background Art

[0002] Compared with optical remote sensing, which is easily affected by bad weather such as clouds, rain, and haze, it is difficult to obtain real-time crop distribution information and cannot fully monitor the growth cycle of crops. Synthetic aperture radar (SAR) remote sensing technology can work all day and all weather, continuously monitor the growth changes of crops, and has become an important means of crop monitoring. Polarimetric SAR (PolSAR) can use multi-polarization channels to reflect biophysical parameter information such as crop morphology, dielectric constant characteristics, and orientation distribution. At the same time, compared with single-view images that can only reflect the shape and structure information of crops at a certain moment, it is difficult to accurately classify crops with similar characteristics. Multi-temporal polarimetric SAR data can use the unique growth trend and phenological information of crops to increase the distinction between crops and reduce the probability of misclassification. However, in actual crop planting, due to differences in growth conditions and field management, even for the same crop, the specific sowing period of different plots will change the phenological evolution rate, resulting in limited classification accuracy in large-scale and complex test areas. At present, a large number of impressive results have been achieved in crop classification monitoring based on multi-temporal polarimetric SAR data.

[0003] Crop classification methods based on multi-temporal polarimetric SAR data can be divided into two categories:

[0004] (1) The classification method based on feature stacking concatenates the feature vectors of a single time phase into a longer feature vector and classifies them through clustering, machine learning, deep learning and other methods. This method regards the time series features as different bands and can select the optimal band combination to a certain extent, but it ignores the temporal correlation between bands and does not use the time-varying feature information of polarimetric SAR.

[0005] (2) The method based on the similarity of time series curves can regard the time series features as time-varying characteristic curves, and classify them by calculating the similarity of the curves between the objects to be identified and the known standard objects. The higher the similarity, the greater the possibility that the objects belong to the same category. At present, the classification method based on time series curve matching has shown great advantages in dealing with the phenological shift problem of crops. Among them, the TWshapeDTW algorithm can obtain the distance similarity between curves through nonlinear mapping from one sequence to another based on the local shape and time constraints of the curve. It has been proven that it can effectively reduce the matching error between curves and improve the accuracy of crop mapping in small sample conditions.

[0006] However, the above methods mostly search for the optimal matching path based on a single time series feature, without considering the consistency constraints of the actual crop phenological changes. The matching results are not robust and are easily affected by outliers. At the same time, the traditional method only uses the cumulative distance sum as a similarity measure, which fails to fully utilize the matching information and crop phenological information, and the distinction between categories needs to be improved. Summary of the invention

[0007] In response to the phenological shift problem of crops, the present invention provides a polarimetric SAR crop classification method with multi-feature joint time series matching, which makes full use of the time-varying characteristics and phenological information of crops, improves the category distinction of crops, and obtains more accurate classification results.

[0008] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0009] A polarimetric SAR crop classification method based on multi-feature joint time series matching, comprising:

[0010] Step 1: obtain a multi-phase polarimetric SAR data set of the crop area, extract multiple features to obtain each feature time series, and calculate the discrimination degree of each polarization feature;

[0011] Step 2: weight the Euclidean distance with the discrimination degree of each feature as the weight, and use it as the base distance of the algorithm for aligning the feature time series curves, and match the path of the feature time series curves of the crop to be classified with the feature time series curves of various standard crop samples;

[0012] Step 3, the matched characteristic time series curves of the crops to be classified are divided into subsequences according to the phenological calendars of various standard crops, and the similarity of each phenological subsequence between the crops to be classified and each standard crop is calculated; then, according to the similarity of the phenological subsequences of various characteristics, the distance similarity between the crops to be classified and each standard crop based on each characteristic is calculated;

[0013] Step 4, using the distinguishing degree of each feature as a weight, weighted calculation is performed on the distance similarity between various features and each standard crop, and the final matching distance similarity between the crop to be classified and each standard crop is obtained; the category of the standard crop with the smallest final matching distance similarity is determined as the category of the crop to be classified.

[0014] Furthermore, the polarization features extracted in step 1 include backscattering coefficient, NNED decomposed power value, total power span and radar vegetation index.

[0015] Furthermore, the calculation method of polarization feature discrimination in step 1 is as follows: first, polarization SAR is segmented into superpixels to obtain a feature time series set of crop category-level objects; then, based on the feature time series set, the discrimination of each polarization feature is calculated according to the following formula:

[0016]

[0017] In the formula, is the discrimination of polarization feature f; is the characteristic time series of the kth sample belonging to crop category c in the polarimetric SAR dataset. The characteristic time series of any sample is X = {x1, x2, ..., x t}, where x i represents the polarization eigenvalue of phase i; is the mean feature time series of all samples of category c; n c is the number of samples included in category c, l is the number of crop categories; d TS is the alignment distance of the timing curve, d ED is the Euclidean distance.

[0018] Furthermore, step 2 specifically includes:

[0019] First, based on the feature discrimination obtained in step 1, a multi-feature joint weighted Euclidean distance is constructed as the distance between different phases of different curves:

[0020]

[0021]

[0022] In the formula, is the discrimination of polarization feature f, W f is the weight of polarization feature f, and F is the number of polarization feature types extracted; The time series curve Af represents the value of the polarization characteristic f of the standard crop sample A at phase i, The time series curve B represents the polarization feature f of the crop sample B to be classified f The value at phase j; d ij For curve A f At time i, the curve B f Multi-feature weighted Euclidean distance at phase j;

[0023] Then, add the dynamic time weight constraint w ij , calculate curve A f At time i, the curve B f The distance d′ at phase j ij :

[0024] d′ij =(w ij +d ij )

[0025]

[0026] In the formula, g is the gain factor; c = |ij| is the time distance factor, m c is the time corresponding to the middle node of the time series curve of the standard crop sample A; ||·||2 is the 2-norm;

[0027] Finally, with the distance d′ ij For the base distance as the characteristic time series curve alignment algorithm, the characteristic time series curve of the crop to be classified is path-matched with the characteristic time series curves of various standard crop samples.

[0028] Furthermore, the calculation method of the phenological subsequence similarity in step 3 is:

[0029] (a1) Assume that the phenological calendar of a standard crop A is: seedling period, 1 to r; growth period, r+1 to p; development period, p+1 to q; maturity period, q+1 to g;

[0030] Then the characteristic time series A of any polarization feature f of standard crop A is f The subsequences are divided into:

[0031]

[0032] The feature time series B of the crop B to be classified f , according to the path matched in step 2, the subsequences are divided by phenology as follows:

[0033]

[0034] In the formula, Respectively represent the subsequences of polarization characteristics f of standard crop A and crop B to be classified in phenological period j, j = 1, 2, 3, 4 correspond to the seedling stage, growth stage, development stage, and maturity stage respectively;

[0035] (a2) Calculate the phenological subsequence similarity between the standard crop A and the crop B to be classified:

[0036]

[0037]

[0038] In the formula, represents the phenological subsequence similarity of the polarization feature f of the standard crop A and the crop to be classified B in phenological period j, d ED() indicates calculating the Euclidean distance.

[0039] Furthermore, the distance similarity between the crop to be classified and each standard crop in step 3 is calculated as follows:

[0040] (b1) Calculate each phenological subsequence The characteristic mean of and

[0041]

[0042]

[0043] In the formula, and Phenological subsequences and The characteristic mean of is the characteristic average value of all phases of each subsequence;

[0044] (b2) The similarity of phenological subsequences As the corresponding subsequence penalty term, according to the feature mean of the phenological subsequence, the distance similarity between the standard crop A and the crop B to be classified based on each phenological polarization feature f is calculated.

[0045]

[0046] In the formula, is the subsequence penalty term of the polarization feature f,

[0047] Furthermore, in step 4, the distance similarity between various features and each standard crop is weighted to obtain the final matching distance similarity d AB The calculation method is:

[0048]

[0049] Where, f represents the fth polarization feature extracted, F is the number of polarization feature types extracted; W f is the discrimination of polarization feature f, is the distance similarity between the standard crop A and the crop B to be classified based on the polarization feature f.

[0050] A system based on any of the above crop classification methods, comprising:

[0051] The feature extraction and discrimination calculation module is used to: obtain the multi-phase polarimetric SAR data set of the crop area, extract multiple features to obtain the time series of each feature, and calculate the discrimination of each polarization feature;

[0052] The path matching module is used to: weight the Euclidean distance with the discrimination degree of each feature as the weight, and use it as the base distance of the algorithm for aligning the feature time series curve, and perform path matching between the feature time series curve of the crop to be classified and the feature time series curve of various standard crop samples;

[0053] The feature distance similarity calculation module is used to: divide the matched feature time series curve of the crop to be classified into subsequences according to the phenological calendars of various standard crops, and calculate the similarity of each phenological subsequence between the crop to be classified and each standard crop; and then calculate the distance similarity based on each feature between the crop to be classified and each standard crop according to the similarity of the phenological subsequences of various features;

[0054] The matching distance similarity calculation and category determination module is used to: use the distinguishing degree of each feature as a weight to perform weighted calculation on the distance similarity between various features and each standard crop, and obtain the final matching distance similarity between the crop to be classified and each standard crop; and determine the category of the standard crop with the smallest final matching distance similarity as the category of the crop to be classified.

[0055] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor implements any one of the above-mentioned crop classification methods.

[0056] A computer-readable storage medium stores a computer program, wherein the computer program implements any of the above-mentioned crop classification methods when executed by a processor.

[0057] Beneficial Effects

[0058] In response to the problem of crop phenological shift, the present invention constructs a polarimetric SAR crop classification method with multi-feature joint time series matching, which can effectively solve the problems of phenological cycle changes and time distortion caused by meteorology, planting activities, etc., and is more suitable for crop classification mapping in small samples, large areas or complex research areas.

[0059] On the one hand, the present invention considers the consistency constraint of phenological changes, reduces the influence of outliers through multi-feature joint weighted time series matching, and obtains more robust matching results; on the other hand, it constructs feature subsequences based on the crop phenological calendar and performs subsequence matching classification, so as to make fuller use of the time-varying characteristics and phenological information of crops and improve the classification distinction of crops. It provides a reliable method for improving the accuracy of crop classification mapping based on time series feature curve matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1is a flow chart of the method described in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The following is a detailed description of an embodiment of the present invention. This embodiment is based on the technical solution of the present invention, and provides a detailed implementation method and a specific operation process to further explain the technical solution of the present invention.

[0062] In order to more clearly illustrate the purpose and implementation process of the present invention, this embodiment selects the farmland dataset of the Flevoland experimental area of ​​the AgriSAR 2009 project to further illustrate the method proposed in the present invention. It should be understood that the specific implementation described herein is only used to explain the present invention and is not used to limit the present invention.

[0063] Example 1

[0064] This embodiment provides a polarimetric SAR crop classification method based on multi-feature joint time series matching. Figure 1 As shown, the following steps are included:

[0065] Step 1: Multi-phase polarimetric SAR data processing: First, the multi-phase polarimetric SAR data is registered and filtered, and then polarimetric feature extraction and polarimetric SAR superpixel segmentation are performed to obtain the object-level temporal polarimetric feature dataset; then, the feature separation index DFS is used TS Calculate the feature discrimination and evaluate the ability of different time series polarization features to distinguish crop categories. Specifically:

[0066] (1) Registration and filtering: One scene in the time-series polarimetric SAR image is selected as the reference image, the remaining images are registered to the coordinate system of the reference image, and then the time-series images are subjected to speckle noise filtering.

[0067] (2) Time series feature extraction: Polarimetric decomposition is performed on the polarimetric SAR image to extract pixel-level polarimetric features. Multi-phase features are connected in series in time to obtain a time series of polarimetric features. The time series features are then normalized. The polarimetric features used in this embodiment include: backscatter coefficients: SigmaHV, SigmaHH, SigmaVV; NNED decomposition power values: Ps, Pd, Pv, and total power Span; Radar vegetation index: RVI; Cloude decomposition feature parameters: H, A, Alpha.

[0068] (3) Polarimetric SAR superpixel segmentation: Perform SLIC segmentation on multi-temporal images to obtain superpixel segmentation results.

[0069] (4) Generate object-level temporal polarization feature dataset: Average the polarization features of the pixels contained in each superpixel block to obtain a crop category-level and object-level temporal polarization feature dataset.

[0070] (5) Time series feature optimization: Building DFS based on time series matching algorithm Ts The feature separation index calculates the discrimination of different temporal polarization features and is used to evaluate the ability of each feature to distinguish crop categories.

[0071]

[0072] In the formula, is the discrimination of polarization feature f; is the characteristic time series of the kth sample belonging to crop category c in the polarimetric SAR dataset. The characteristic time series of any sample is X = {x1, x2, ..., x t}, where x i represents the polarization eigenvalue of phase i; is the mean feature time series of all samples of category c; n c is the number of samples included in category c, l is the number of crop categories; d Ts is the alignment distance of the timing curve, d ED is the Euclidean distance.

[0073] Step 2, multi-feature joint weighted time series matching: First, the feature discrimination obtained in the step is used as the weight to weight the polarization feature time series, and then the multi-feature joint weighted Euclidean distance is used to replace the traditional single feature Euclidean distance as the base distance of the time series curve alignment algorithm to match the time series feature curve. Specifically:

[0074] Assume that the time series curve of an arbitrary polarization feature f of a standard crop sample A is The time series curve of the polarization feature f of the crop sample B to be classified is:

[0075] First, based on the feature discrimination obtained in step 1, a multi-feature joint weighted Euclidean distance is constructed to obtain the distance matrix d between each time node. m×n , whose matrix elements can be expressed as:

[0076]

[0077]

[0078] In the formula, is the discrimination degree of polarization feature f, Wf is the weight of polarization feature f, and F is the number of polarization feature types extracted; The time series curve A representing the polarization characteristic ff The value at phase i, The time series curve B showing the polarization characteristic f f The value at phase j; d ij is the multi-feature weighted Euclidean distance between curve A at phase i and curve B at phase j.

[0079] Then, add the dynamic time weight constraint w ij , calculate the distance d′ between curve A at phase i and curve B at phase j ij :

[0080] d′ ij =(w ij +d ij )

[0081]

[0082] Where g is the gain factor; c = |ij| is the time distance factor, mc is the time corresponding to the middle node of the time series curve of the standard crop sample A; ||·||2 is the 2-norm; where the parameters g and m c The optimal solution can be determined based on the training samples through 5-fold cross validation.

[0083] Finally, with the distance d′ ij For the base distance as the characteristic time series curve alignment algorithm, the characteristic time series curve of the crop to be classified is path-matched with the characteristic time series curves of various standard crop samples.

[0084] The path matching in this embodiment is implemented by using the existing technology: based on the constraints of boundary, continuity and monotonicity, the cumulative distance matrix Q is calculated by recursively accumulating the minimum distance m×n , whose matrix elements can be expressed as:

[0085] Q ij =min{Q i-1，j , Q i-1，j-1 , Q i，j-1 )+d′ ij Q 11 =d′ 11

[0086] In the formula, i and j are the distance matrix Q m×n The row and column number of a certain position in ;

[0087] Based on the path and minimum constraints, the optimal matching path P can be obtained using a dynamic search algorithm, which can be expressed as:

[0088]

[0089] For the crop B to be classified in the present invention, due to the problem of phenological uncertainty, in the i phase, the standard crop A may grow to flowering, and the crop B may have already borne fruit. In this case, the characteristics of the two are not corresponding, and the information reflected is different. Therefore, by constructing an algorithm to achieve path matching, the period of B corresponding to each period of A can be found. For example, the period corresponding to the i phase (flowering) of B and A may be the i-1 period. Then, when calculating the curve distance similarity classification, the distance between the i period feature of A and the i-1 period feature of B is used.

[0090] Step 3, feature subsequence division: based on the phenological calendar of each standard crop, divide the polarization feature time series of the crop to be classified into subsequences; and calculate the similarity of each phenological subsequence between the crop to be classified and each standard crop; then, based on the similarity of the phenological subsequences of various features, calculate the distance similarity between the crop to be classified and each standard crop based on each feature. Specifically:

[0091] First, assuming that the phenological calendar of a standard crop A is: seedling period, 1 to r; growth period, r+1 to p; development period, p+1 to q; maturity period, q+1 to g; then the feature time series Af of any polarization feature f of the standard crop A is divided into subsequences according to phenology:

[0092]

[0093] The characteristic time series Bf of the crop B to be classified is divided into subsequences according to the path matched in step 2:

[0094]

[0095] In the formula, They represent the subsequences of the polarization characteristics f of the standard crop A and the crop B to be classified in the phenological period j, respectively, and j=1, 2, 3, 4 correspond to the seedling period, the growth period, the development period, and the maturity period, respectively.

[0096] Then, the phenological subsequence similarity between the standard crop A and the crop B to be classified is calculated:

[0097]

[0098]

[0099] In the formula, represents the phenological subsequence similarity of the polarization feature f of the standard crop A and the crop to be classified B in phenological period j, d ED () indicates calculating the Euclidean distance.

[0100] Finally, the distance similarity between the crop to be classified and each standard crop is calculated using the following method:

[0101] (1) Calculate each phenological subsequence The characteristic mean of and

[0102]

[0103]

[0104] In the formula, and Phenological subsequences and The characteristic mean of is the characteristic average value of all phases of each subsequence;

[0105] (2) The similarity of phenological subsequences As the corresponding subsequence penalty term, according to the feature mean of the phenological subsequence, the distance similarity between the standard crop A and the crop B to be classified based on each phenological polarization feature f is calculated.

[0106]

[0107] In the formula, is the subsequence penalty term of the polarization feature f,

[0108] Step 4, using the distinguishing degree of each feature as a weight, weighted calculation is performed on the distance similarity between various features and each standard crop, and the final matching distance similarity between the crop to be classified and each standard crop is obtained; the category of the standard crop with the smallest final matching distance similarity is determined as the category of the crop to be classified.

[0109] Among them, the distance similarity between various features and each standard crop is weighted to obtain the final matching distance similarity d AB The calculation method is:

[0110]

[0111] Where, f represents the fth polarization feature extracted, F is the number of polarization feature types extracted; W f is the discrimination of polarization feature f, is the distance similarity between the standard crop A and the crop B to be classified based on the polarization feature f.

[0112] Example 2

[0113] This embodiment provides a polarimetric SAR crop classification system based on multi-feature joint time series matching, including the following modules:

[0114] The feature extraction and discrimination calculation module is used to: obtain the multi-phase polarimetric SAR data set of the crop area, extract multiple features to obtain the time series of each feature, and calculate the discrimination of each polarization feature;

[0115] The path matching module is used to: weight the Euclidean distance with the discrimination degree of each feature as the weight, and use it as the base distance of the algorithm for aligning the feature time series curve, and perform path matching between the feature time series curve of the crop to be classified and the feature time series curve of various standard crop samples;

[0116] The feature distance similarity calculation module is used to: divide the matched feature time series curve of the crop to be classified into subsequences according to the phenological calendars of various standard crops, and calculate the similarity of each phenological subsequence between the crop to be classified and each standard crop; and then calculate the distance similarity based on each feature between the crop to be classified and each standard crop according to the similarity of the phenological subsequences of various features;

[0117] The matching distance similarity calculation and category determination module is used to: use the distinguishing degree of each feature as a weight to perform weighted calculation on the distance similarity between various features and each standard crop, and obtain the final matching distance similarity between the crop to be classified and each standard crop; and determine the category of the standard crop with the smallest final matching distance similarity as the category of the crop to be classified.

[0118] The modules included in the crop classification system provided in this embodiment belong to the function implementation modules of the method described in Example 1. The specific working principles and implementation methods are the same as those described in Example 1 and will not be repeated here.

[0119] Example 3

[0120] This embodiment provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor implements the polarization SAR crop classification method based on multi-feature joint time series matching described in Example 1.

[0121] Example 4

[0122] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the polarization SAR crop classification method based on multi-feature joint time series matching described in Example 1.

[0123] The above embodiments are preferred embodiments of the present application. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the technical concept of the embodiments of the present invention, the technical solutions of the embodiments of the present invention can be modified in a variety of simple ways, and these simple modifications all belong to the protection scope of the embodiments of the present invention. In addition, various changes or improvements can be made between the various different implementations of the embodiments of the present invention, which should also be regarded as the contents disclosed in the embodiments of the present invention without departing from the general concept of the present application.

Claims

1. A polarimetric SAR crop classification method based on multi-feature joint time series matching, characterized in that: include: Step 1: obtain a multi-phase polarimetric SAR data set of the crop area, extract multiple features to obtain each feature time series, and calculate the discrimination degree of each polarization feature; Step 2: weight the Euclidean distance with the discrimination degree of each feature as the weight, and use it as the base distance of the algorithm for aligning the feature time series curves, and match the path of the feature time series curves of the crop to be classified with the feature time series curves of various standard crop samples; Step 3, the matched characteristic time series curves of the crops to be classified are divided into subsequences according to the phenological calendars of various standard crops, and the similarity of each phenological subsequence between the crops to be classified and each standard crop is calculated; then, according to the similarity of the phenological subsequences of various characteristics, the distance similarity between the crops to be classified and each standard crop based on each characteristic is calculated; Step 4, using the distinguishing degree of each feature as a weight, weighted calculation is performed on the distance similarity between various features and each standard crop, and the final matching distance similarity between the crop to be classified and each standard crop is obtained; the category of the standard crop with the smallest final matching distance similarity is determined as the category of the crop to be classified.

2. The crop classification method according to claim 1, characterized in that: The polarization features extracted in step 1 include the backscattering coefficient, the power value of NNED decomposition, the total power span, and the radar vegetation index.

3. The crop classification method according to claim 1, characterized in that: The calculation method of polarization feature discrimination in step 1 is as follows: first, polarization SAR is segmented into superpixels to obtain a feature time series set of crop category-level objects; then, based on the feature time series set, the discrimination of each polarization feature is calculated according to the following formula: In the formula, is the discrimination of polarization feature f; is the characteristic time series of the kth sample belonging to crop category c in the polarimetric SAR dataset. The characteristic time series of any sample is X = {x1, x2, ..., x t }, where x i represents the polarization eigenvalue of phase i; is the mean feature time series of all samples of category c; n c is the number of samples included in category c, l is the number of crop categories; d TS is the alignment distance of the timing curve, d ED is the Euclidean distance.

4. The crop classification method according to claim 1, characterized in that: Step 2 specifically includes: First, based on the feature discrimination obtained in step 1, a multi-feature joint weighted Euclidean distance is constructed as the distance between different phases of different curves: In the formula, is the discrimination of polarization feature f, W f is the weight of polarization feature f, and F is the number of polarization feature types extracted; The time series curve A represents the polarization characteristic f of the standard crop sample A f The value at phase i, The time series curve B represents the polarization feature f of the crop sample B to be classified f The value at phase j; d ij For curve A f At phase i and curve B f Multi-feature weighted Euclidean distance at phase j; Then, add the dynamic time weight constraint w ij , calculate curve A f At phase i and curve B f The distance d' at phase j ij : d′ ij =(w ij +d ij ) In the formula, g is the gain factor; c = |ij| is the time distance factor, m c is the time corresponding to the middle node of the time series curve of the standard crop sample A; ||·||2 is the 2-norm; Finally, with a distance d' ij For the base distance as the characteristic time series curve alignment algorithm, the characteristic time series curve of the crop to be classified is path-matched with the characteristic time series curves of various standard crop samples.

5. The crop classification method according to claim 1, characterized in that: The calculation method of phenological subsequence similarity in step 3 is: (a1) Assume that the phenological calendar of a standard crop A is: seedling period, 1 to r; growth period, r+1 to p; development period, p+1 to q; maturity period, q+1 to g; Then the characteristic time series A of any polarization feature f of standard crop A is f The subsequences are divided into: The feature time series B of the crop B to be classified f , according to the path matched in step 2, the subsequences are divided by phenology as follows: In the formula, Respectively represent the subsequences of polarization characteristics f of standard crop A and crop B to be classified in phenological period j, j = 1, 2, 3, 4 correspond to the seedling stage, growth stage, development stage, and maturity stage respectively; (a2) Calculate the phenological subsequence similarity between the standard crop A and the crop B to be classified: In the formula, represents the similarity of the phenological subsequences of the polarization feature f of the standard crop A and the crop B to be classified in the phenological period j, d ED () indicates calculating the Euclidean distance.

6. The crop classification method according to claim 5, characterized in that: The distance similarity between the crop to be classified and each standard crop described in step 3 is calculated as follows: (b1) Calculate each phenological subsequence The characteristic mean of and In the formula, and Phenological subsequences and The characteristic mean of is the characteristic average value of all phases of each subsequence; (b2) The similarity of phenological subsequences As the corresponding subsequence penalty term, according to the feature mean of the phenological subsequence, the distance similarity between the standard crop A and the crop B to be classified based on each phenological polarization feature f is calculated. In the formula, is the subsequence penalty term of the polarization feature f, 7. The crop classification method according to claim 1, characterized in that: In step 4, the distance similarity between various features and each standard crop is weighted to obtain the final matching distance similarity d AB The calculation method is: In the formula, f represents the fth polarization feature extracted, and F is the number of polarization feature types extracted; W f is the discrimination of polarization feature f, is the distance similarity between the standard crop A and the crop B to be classified based on the polarization feature f.

8. A system based on the method according to any one of claims 1 to 7, characterized in that: include: The feature extraction and discrimination calculation module is used to: obtain the multi-phase polarimetric SAR data set of the crop area, extract multiple features to obtain the time series of each feature, and calculate the discrimination of each polarization feature; The path matching module is used to: weight the Euclidean distance with the discrimination degree of each feature as the weight, and use it as the base distance of the algorithm for aligning the feature time series curve, and perform path matching between the feature time series curve of the crop to be classified and the feature time series curve of various standard crop samples; The feature distance similarity calculation module is used to: divide the matched feature time series curve of the crop to be classified into subsequences according to the phenological calendars of various standard crops, and calculate the similarity of each phenological subsequence between the crop to be classified and each standard crop; and then calculate the distance similarity based on each feature between the crop to be classified and each standard crop according to the similarity of the phenological subsequences of various features; The matching distance similarity calculation and category determination module is used to: use the distinguishing degree of each feature as a weight to perform weighted calculation on the distance similarity between various features and each standard crop, and obtain the final matching distance similarity between the crop to be classified and each standard crop; and determine the category of the standard crop with the smallest final matching distance similarity as the category of the crop to be classified.

9. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the computer program is executed by the processor, the processor is caused to implement the method according to any one of claims 1 to 7.

10. 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 according to any one of claims 1 to 7 is implemented.

Images