An early classification method and system for overwintering crops based on enhanced phenological characteristics

Abstract

The present application relates to a method or apparatus for data recognition using electronic devices, providing a method and system for early classification of overwintering crops based on enhanced phenological characteristics. The method comprises: determining, based on pre-acquired time-series remote sensing images, multiple categories of time-series classification feature images corresponding to multiple phenological periods in the early growth period of a target overwintering crop; performing feature screening on the multiple categories of time-series classification feature images to determine a set of classification feature variables for each phenological period; determining a set of enhanced phenological characteristic indices for each phenological period based on the set of classification feature variables for each phenological period; and training a pre-constructed first crop classification model based on the enhanced phenological characteristic indices to classify the target overwintering crop based on the trained first crop classification model. In this way, by enhancing crop classification features based on phenological characteristics, the accuracy of early recognition of overwintering crops is improved.

Description

Technical Field

[0001] The present application relates to the technical field of methods or devices for data identification using electronic devices, and in particular to a method and system for early classification of overwintering crops based on enhanced phenological characteristics. Background Art

[0002] In the early growth stages of overwintering crops (such as winter wheat), the leaves of the crops have not yet grown and cannot completely cover the ground surface. Part of the soil in the planting area is still exposed, resulting in the mixing of vegetation information and soil information, and weak vegetation characteristics, which brings a series of challenges to the early identification of overwintering crops.

[0003] Therefore, it is necessary to provide an improved technical solution to the above-mentioned deficiencies in the prior art. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for early classification of overwintering crops based on enhanced phenological characteristics, so as to solve or alleviate the problems existing in the above-mentioned prior art.

[0005] In order to achieve the above objectives, this application provides the following technical solutions:

[0006] The present application provides an early classification method for overwintering crops based on enhanced phenological characteristics, including:

[0007] Based on pre-acquired time-series remote sensing images, multiple categories of time-series classification feature images corresponding to multiple phenological periods in the early growth stage of target overwintering crops are determined; wherein the multiple phenological periods in the early growth stage of the target overwintering crops include at least a first phenological period and a second phenological period, the first phenological period represents the sowing period of the target overwintering crops, and the second phenological period represents the early overwintering period of the target overwintering crops or the greening period of the target overwintering crops; the multiple categories of time-series classification feature images include at least phenological change feature images;

[0008] Performing feature screening on the multiple categories of time-series classification feature images to determine a set of classification feature variables for each of the phenological periods;

[0009] Determining a set of enhanced phenological characteristic indices for each phenological period based on a set of classification characteristic variables for each phenological period;

[0010] A pre-constructed first crop classification model is trained according to the enhanced phenological characteristic index set, so as to classify the target overwintering crops based on the trained first crop classification model to obtain a classification result of the target overwintering crops.

[0011] Preferably, the feature screening of the multiple categories of time-series classification feature images to determine the classification feature variable set of each phenological period includes:

[0012] Calculating, based on a random forest method, an importance score of each classification feature in the time-series classification feature images of the multiple categories of each phenological period, and determining a first classification feature variable set based on the importance score of each classification feature;

[0013] Based on the analysis method of correlation analysis, the first classification characteristic variable set is screened for correlation to determine the classification characteristic variable set of each phenological period.

[0014] Preferably, the step of calculating the importance score of each classification feature in the time-series classification feature images of the multiple categories of each phenological period based on the random forest method, and determining the first classification feature variable set based on the importance score of each classification feature, includes:

[0015] Step S111, determining the importance score of each classification feature in the time series classification feature images of the multiple categories in each phenological period based on the out-of-bag error of the random forest method;

[0016] Step S112: sort the importance scores in descending order, and remove the classification features corresponding to the M importance scores with the lowest order, to obtain a new classification feature variable subset; wherein M is an integer greater than 1;

[0017] Step S113: determining the classification accuracy of the target overwintering crop based on the new classification feature variable subset;

[0018] Step S114, iteratively executing steps S111 to S113 until the classification accuracy of the target overwintering crops converges, and using the classification feature variable subset corresponding to the convergence of the classification accuracy of the target overwintering crops as the first classification feature variable set.

[0019] Preferably, the determining of the enhanced phenological characteristic index set for each phenological period based on the classification characteristic variable set for each phenological period includes:

[0020] The time series classification feature images corresponding to each classification feature variable in the classification feature variable set of each phenological period are synthesized to obtain an enhanced phenological feature index set of each phenological period; wherein the classification feature variables in the classification feature variable set correspond one-to-one to the enhanced phenological feature indices in the enhanced phenological feature index set.

[0021] Preferably, the phenological change characteristics include a crop phenological difference index; the crop phenological difference index is calculated according to the following formula:

[0022] WPDI=Max{NDPI Pho_2}-Min{NDPI Pho_1}

[0023] Wherein, WPDI represents the crop phenological difference index; NDPI Pho_1 The time series data of the normalized phenological index of the first phenological period, NDPI Pho_2 The time series data representing the normalized phenological index of the second phenological period.

[0024] Preferably, the phenological change characteristics include a crop phenological ratio index; the crop phenological ratio index is calculated according to the following formula:

[0025]

[0026] Wherein, WPRI represents the crop phenology ratio index; NDPI Pho_1 The time series data of the normalized phenological index of the first phenological period, NDPI Pho_2 The time series data representing the normalized phenological index of the second phenological period.

[0027] Preferably, the phenological change characteristics further include the annual accumulated days corresponding to the maximum value of the normalized phenological index in the pre-wintering time window;

[0028] According to the formula:

[0029] DOY max =argmax{NDPI pho |pho _3 ≤pho≤pho _4}

[0030] Determining the annual accumulated day corresponding to the maximum value of the normalized phenological index within the pre-wintering time window;

[0031] Where, DOY max The annual cumulative days corresponding to the maximum value of the normalized phenological index in the pre-wintering time window; NDPI pho represents the normalized phenological index; pho_3 represents the sowing period of the target overwintering crop; pho_4 represents the early overwintering period of the target overwintering crop.

[0032] Preferably, before training the pre-constructed first crop classification model according to the enhanced phenological characteristic index set, the method further comprises:

[0033] The hyperparameters of the pre-built first crop classification model are tuned by a grid search algorithm.

[0034] Preferably, the method further comprises:

[0035] Based on the pre-built second crop classification model, other overwintering crops other than the target overwintering crop in the same period are classified to obtain classification results of other overwintering crops;

[0036] Superimposing the classification result of the target overwintering crop with the classification results of the other overwintering crops to determine an overlapping area between the classification result of the target overwintering crop and the classification results of the other overwintering crops;

[0037] The overlapping area is removed from the classification results of the target overwintering crops.

[0038] The present application provides an early classification system for overwintering crops based on enhanced phenological characteristics, including:

[0039] The first computing unit is configured to determine, based on pre-acquired time-series remote sensing images, a plurality of categories of time-series classification characteristic images corresponding to a plurality of phenological periods in the early growth stage of a target overwintering crop, wherein the plurality of phenological periods in the early growth stage of the target overwintering crop include at least a first phenological period and a second phenological period, the first phenological period representing the sowing period of the target overwintering crop, and the second phenological period representing the early overwintering period of the target overwintering crop or the greening period of the target overwintering crop; the time-series classification characteristic images include at least phenological change characteristic images;

[0040] a feature screening unit configured to perform feature screening on the time series classification feature images of the multiple categories to determine a set of classification feature variables for each of the phenological periods;

[0041] a second calculation unit configured to determine a set of enhanced phenological characteristic indices for each phenological period based on a set of classification characteristic variables for each phenological period;

[0042] The crop classification unit is configured to train a pre-constructed first crop classification model according to the enhanced phenological characteristic index set, so as to classify the target overwintering crops based on the trained first crop classification model to obtain a classification result of the target overwintering crops.

[0043] Beneficial effects:

[0044] Through the above scheme, it can be seen that the present application provides an early classification method for overwintering crops based on enhanced phenological characteristics, including: based on pre-acquired time-series remote sensing images, determining multiple categories of time-series classification feature images corresponding to multiple phenological periods in the early growth stage of the target overwintering crops; performing feature screening on multiple categories of time-series classification feature images to determine a set of classification feature variables for each phenological period; based on the set of classification feature variables for each phenological period, determining a set of enhanced phenological feature indexes for each phenological period; training a pre-constructed first crop classification model according to the enhanced phenological feature index, and classifying the target overwintering crops based on the trained first crop classification model. In this way, after acquiring the time-series remote sensing images, according to the multiple phenological periods in the early growth stage of the target overwintering crops, the multiple categories of time-series classification feature images corresponding to each phenological period are determined respectively, and the classification feature images for different phenological periods are different, thereby ensuring that the classification features are adapted to the phenological period of the crop to the greatest extent. Among them, multiple categories of time series classification feature images include at least phenological change feature images, so as to enhance the vegetation characteristics of overwintering crops in the early stage due to their weak characteristics, thereby improving the recognition accuracy of overwintering crops in the early stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings and descriptions that constitute part of this application are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. Among them:

[0046] Figure 1 A schematic flow chart of an early classification method for overwintering crops based on enhanced phenological characteristics according to some embodiments of the present application;

[0047] Figure 2 A logical diagram of an early classification method for overwintering crops based on enhanced phenological characteristics according to some embodiments of the present application;

[0048] Figure 3 A schematic diagram of time windows corresponding to different early phenological phases of overwintering crops according to some embodiments of the present application;

[0049] Figure 4 A schematic diagram of a process for obtaining a first classification feature variable set according to some embodiments of the present application;

[0050] Figure 5 A schematic diagram of parameter tuning results of an SVM according to some embodiments of the present application;

[0051] Figure 6 A schematic diagram of parameter tuning results of the random forest algorithm RF provided according to some embodiments of the present application;

[0052] Figure 7A schematic diagram of parameter tuning results of the GBDT algorithm provided according to some embodiments of the present application;

[0053] Figure 8 A probability map of early identification results of overwintering crops provided according to some embodiments of the present application;

[0054] Figure 9 A schematic diagram showing a comparison between the early identification results of overwintering crops in region A provided according to some embodiments of the present application and the statistical data on the area of the overwintering crops released by the statistical department;

[0055] Figure 10 A schematic diagram showing a comparison between the early identification results of overwintering crops in region B provided according to some embodiments of the present application and the statistical data on the area of the overwintering crops released by the statistical department;

[0056] Figure 11 A schematic diagram showing a comparison between the early identification results of overwintering crops in Region C provided according to some embodiments of the present application and the statistical data on the area of overwintering crops released by the statistical department;

[0057] Figure 12 This is a schematic structural diagram of an early classification system for overwintering crops based on enhanced phenological characteristics according to some embodiments of the present application. DETAILED DESCRIPTION

[0058] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments. Each example is provided by way of explanation of the present application and does not limit the present application. In fact, it will be clear to those skilled in the art that modifications and variations can be made in the present application without departing from the scope or spirit of the present application. For example, a feature shown or described as part of one embodiment can be used in another embodiment to produce yet another embodiment. Therefore, it is expected that the present application includes such modifications and variations within the scope of the appended claims and their equivalents.

[0059] Exemplary Methods

[0060] The present invention provides an early classification method for overwintering crops based on enhanced phenological characteristics. Figure 1 A schematic flow chart of an early classification method for overwintering crops based on enhanced phenological characteristics according to some embodiments of the present application; Figure 2 This is a logical diagram of an early classification method for overwintering crops based on enhanced phenological characteristics according to some embodiments of the present application. Figure 1 、 Figure 2 As shown, the method includes:

[0061] Step S101: Based on the pre-acquired time-series remote sensing images, determine multiple categories of time-series classification feature images corresponding to multiple phenological periods in the early growth stage of the target overwintering crop; wherein the multiple phenological periods in the early growth stage of the target overwintering crop include at least a first phenological period and a second phenological period, the first phenological period represents the sowing period of the target overwintering crop, and the second phenological period represents the early overwintering period of the target overwintering crop or the greening period of the target overwintering crop; the multiple categories of time-series classification feature images include at least phenological change feature images.

[0062] In the examples of this application, overwintering crops, also known as autumn sowing, refer to crops that are sown in autumn, whose seedlings survive the winter, and are harvested in the spring or summer of the following year, such as winter wheat, rapeseed, and garlic. This example of the application uses winter wheat as the target overwintering crop to be classified to illustrate the technical solution.

[0063] It should be noted that the early growth period of overwintering crops can be the period from the time the crops are sown in autumn and pass through the winter to the time they turn green in the spring of the following year.

[0064] The phenological periods corresponding to the early growth period of overwintering crops may include: sowing period (first phenological period), early overwintering period, and greening period in the second year (second phenological period).

[0065] In an embodiment of the present application, a time series curve is drawn based on the Normalized Difference Phenology Index NDPI (Normalized Difference Phenology Index) of the target overwintering crop, and the time windows corresponding to different phenological periods in the early growth stage of the target overwintering crop are determined by analyzing the NDPI index time series curve.

[0066] Specifically, the time windows corresponding to different phenological phases of early winter wheat are as follows: Figure 3 As shown. Combined Figure 3Analysis shows that during the sowing period, winter wheat planting areas have low chlorophyll content, resulting in the lowest NDPI values for winter wheat. During this period, the NDPI values for woodlands and grasslands are higher than those for winter wheat, the NDPI values for impervious surfaces are similar to those for winter wheat, and the NDPI values for water bodies are negative. Classifying winter crops during the sowing period based solely on the spectral band characteristics of remote sensing imagery or by extracting spectral index features from remote sensing imagery can easily lead to misclassification of winter wheat planting areas and bare land. During the early overwintering period, after germination and growth, winter wheat increases its chlorophyll content and NDPI value, typically exceeding 0.35, and the first peak appears in the winter wheat NDPI time series curve. Meanwhile, chlorophyll content decreases in areas such as woodlands and grasslands, and their NDPI values drop to below 0.2. Therefore, the time window for the early overwintering period is defined as the period when the winter wheat NDPI value is greater than 0.35 and less than 0.2. This time window can be used to distinguish winter wheat from areas such as woodlands and grasslands. When the NDPI value of winter wheat increases until a second peak appears, and this peak is higher than the first peak of the NDPI time series curve during the early wintering period, this period is identified as the greening period of winter wheat. During this period, woodlands and grasslands have just entered the greening period, and their NDPI values are low. Impervious surfaces remain essentially unchanged, while the NDPI value of water bodies increases slightly but remains below 0.1. This effectively distinguishes winter wheat from woodlands, grasslands, impervious surfaces, and water bodies.

[0067] From the above analysis, it can be seen that the different phenological periods in the early growth period of winter wheat are different from the phenological periods used to distinguish winter wheat from other surface types (such as bare land, water bodies, woodlands, grasslands, etc.) for crop classification. In addition, in addition to the spectral band characteristics and spectral index characteristics of remote sensing images, phenological change characteristics are crucial for crop classification.

[0068] In an embodiment of the present application, based on pre-acquired time-series remote sensing images, time-series classification feature images of multiple categories corresponding to multiple phenological stages in the early growth stage of the target overwintering crop are determined.

[0069] In practice, time-series remote sensing images are acquired from the Google Earth Engine (GEE) platform and preprocessed for each image. Preprocessing can include noise reduction and cloud removal, such as removing cloud contamination from image data using a cloud mask.

[0070] The acquired time series remote sensing images can include time series remote sensing data from multiple years. For example, you can filter out Sentinel-2 time series image data for the winter wheat growth stages of 2018, 2019, 2020, 2021, and 2022 from the GEE platform.

[0071] Based on the time windows corresponding to the different phenological phases of overwintering crops (i.e., sowing period, early overwintering period, and greening period), multiple categories of time series classification feature images are calculated within the different phenological phase time windows. The multiple categories of time series classification feature images include at least phenological change feature images.

[0072] In some embodiments, the phenological change characteristic includes a crop phenological difference index; the crop phenological difference index is calculated according to the following formula:

[0073] WPDI=Max{NDPI Pho_2}-Min{NDPI Pho_1}

[0074] Wherein, WPDI represents the crop phenological difference index; NDPI represents the crop phenological difference index; Pho_1 Represents the time series data of the normalized phenological index of the first phenological period, NDPI Pho_2 Time series data representing the normalized phenological index of the second phenological period.

[0075] In other embodiments, the phenological change characteristics include a crop phenological ratio index; the crop phenological ratio index is calculated according to the following formula:

[0076]

[0077] Wherein, WPRI represents the crop phenology ratio index; NDPI Pho_1 Represents the time series data of the normalized phenological index of the first phenological period, NDPI Pho_2 Time series data representing the normalized phenological index of the second phenological period.

[0078] In practical applications, after obtaining time-series remote sensing images from the GEE (Google Earth Engine) platform in advance, the time-series characteristic images included in the sowing period, early wintering period and greening period are first determined based on the different phenological periods of the early growth of overwintering crops. Then, based on the time-series characteristic images included in the sowing period, early wintering period and greening period, the time-series NDPI index minimum image within the sowing period time window, the time-series NDPI index maximum image within the early wintering period time window and the time-series NDPI index maximum image within the greening period time window are synthesized respectively. The WPDI and WPRI indices of winter wheat are calculated based on the images corresponding to the synthesized NDPI indices.

[0079] It should be noted that if data is missing in the images synthesized according to the time windows of the sowing period, early wintering period, and greening period, the time window corresponding to that phenological period should be expanded until the synthesized imagery fully covers the target area. Data missing here refers to situations where the synthesized imagery does not fully cover the target area. Data missing makes it impossible to calculate the WPDI and WPRI for winter wheat in areas without data.

[0080] In this way, by calculating the WPDI and WPRI of the target overwintering crop (winter wheat), the sensitivity of NDPI to the weaker vegetation characteristics in the early growth stage of winter wheat is fully utilized, thereby highlighting the time series information of winter wheat growth and enhancing the classification characteristics of winter wheat in the early growth stage.

[0081] In another optional embodiment, the phenological change characteristics also include the annual accumulated days corresponding to the maximum value of the normalized phenological index in the early wintering time window.

[0082] According to the formula:

[0083] DOY max =argmax{NDPI pho |pho _3 ≤pho≤pho _4}

[0084] Determine the annual cumulative day corresponding to the maximum value of the normalized phenological index within the pre-wintering time window;

[0085] Where, DOY max The annual cumulative days corresponding to the maximum value of the normalized phenological index in the pre-wintering time window; NDPI pho represents the normalized phenological index; pho_3 represents the sowing period of the target overwintering crop; pho_4 represents the early overwintering period of the target overwintering crop.

[0086] Among them, the annual cumulative day corresponding to the maximum value of the normalized phenological index represents the date when the target wintering crop grows most vigorously in the early growth stage of wintering. max It can assist in distinguishing target overwintering crops from vegetation areas such as woodlands and grasslands.

[0087] In some embodiments, the multiple categories of time-series classification feature images further include: spectral band feature images, spectral index feature images, and terrain feature images.

[0088] In the embodiment of the present application, the temporal classification features of multiple categories within different phenological period time windows are calculated respectively to obtain temporal classification feature images of multiple categories. Among them, the classification features of each category are as follows: 57 spectral band features and spectral index features, 3 phenological change features, and 3 topographic features, resulting in a total of 63 temporal feature images. Detailed information on the classification features can be found in Table 1, which is as follows:

[0089]

[0090]

[0091] Among them, the calculation formula of each characteristic variable in the spectral index feature is as follows:

[0092] BSI=(δ swri1 +δ red )-(δ nir +δ blue ) / (δ swri1 +δ red )+(δ nir +δ blue )

[0093] LSWI=(δ nir +δ swir1 ) / (δ nir -δ swir1 )

[0094] NDSVI=(δ swir1 +δ red ) / (δ swir1 -δ red )

[0095] NDTI=(δ swir1 -δ swir2 ) / (δ swir1 +δ swir2 )

[0096] RENDVI=(δ nir -δ RE2 ) / (δ nir +δ RE2 )

[0097] REP=705+35*(0.5*(δ RE3 +δ red )-δ RE1 ) / (δ RE2 -δ RE1 )

[0098] Where, δ swir1 Indicates the value of the first short-wave infrared band in the remote sensing image; δ redIndicates the value of the red light band in the remote sensing image; δ nir Indicates the value of the near-infrared band in the remote sensing image; δ blue Indicates the value of the blue light band in the remote sensing image; δ swir2 Indicates the value of the second short-wave infrared band in remote sensing images; δ RE2 Represents the value of the second red edge band in the remote sensing image.

[0099] In an embodiment of the present application, the acquired time-series remote sensing images are screened according to the time windows corresponding to different early phenological periods of overwintering crops, and then the time-series classification feature images of different phenological periods are calculated based on the screened time-series remote sensing images, so that the candidate classification feature images of each phenological period include the above-mentioned 63 time-series feature images. In subsequent steps, feature screening will be performed based on the importance of these time-series feature images to crop classification, and classification features with higher contribution degrees for each phenological period will be found to improve the accuracy of crop classification.

[0100] Step S102: Perform feature screening on multiple categories of time-series classification feature images to determine a set of classification feature variables for each phenological period.

[0101] It should be understood that if a classification feature can effectively distinguish the target overwintering crop from other landform types, then the classification feature is considered to be an important classification feature and its contribution to the crop classification results is relatively high. In the embodiment of the present application, the candidate classification features of each phenological period (i.e., 63 temporal features) are screened to determine the classification features with higher importance scores for each phenological period in order to optimize the crop classification features and reduce the input parameter dimension of the crop classification model.

[0102] According to the above analysis of different phenological periods, the crop classification characteristics that can be used to distinguish winter wheat from other surface types (such as bare land, water bodies, woodlands, grasslands, etc.) are not the same in different phenological periods of the early growth of winter wheat, and each crop classification characteristic has a different importance score for the classification results.

[0103] In some embodiments, feature screening is performed on multiple categories of time-series classification feature images to determine a set of classification feature variables for each phenological period, including: calculating the importance score of each classification feature in multiple categories of time-series classification feature images for each phenological period based on a random forest method, and determining a first set of classification feature variables based on the importance score of each classification feature; and performing correlation screening on the first set of classification feature variables based on an analysis method of correlation analysis to determine a set of classification feature variables for each phenological period.

[0104] Here, the Random Forest (RF) method is an algorithm that integrates multiple decision trees through the concept of ensemble learning. Its basic unit is the decision tree, and each decision tree is a crop classifier. For an input training sample, N trees will have N classification results. The classification voting results of all crop classifiers are integrated through the Random Forest method, and the category with the most votes is designated as the final crop category.

[0105] In the application scenarios of related technologies, the random forest method is often used to achieve crop classification and fitting. In the embodiment of the present application, the random forest method is used to screen the time series classification feature images to determine the best classification feature set for each phenological period.

[0106] In specific implementation, the random forest method is first used to calculate the importance score of each classification feature in the time-series classification feature image of multiple categories for each phenological period. This is then used to measure the importance of the classification features and select M classification features with higher importance to obtain the first classification feature variable set, achieving a primary screening of the classification features. Then, a correlation analysis-based analysis method is used to calculate the correlation between any two classification feature variables in the first classification feature variable set. Any one of the two classification feature variables with a correlation greater than a preset correlation threshold is removed, achieving a secondary screening of the classification features.

[0107] Here, the analysis method based on correlation analysis is a statistical analysis method for studying the correlation between two or more random variables of equal status. For example, it can be any one of principal component analysis (PCA), canonical correlation analysis (CCA), kernel canonical correlation analysis (KCCA), and sparse canonical correlation analysis (SCCA). Other statistical analysis methods that can achieve correlation analysis can also be used, and this application does not limit this.

[0108] In this way, the classification features of the time series classification feature images of multiple categories are screened once based on the importance score to obtain the first classification feature variable set, and then the correlation analysis method is used to perform a secondary screening of the classification features to obtain the optimal classification feature set for each phenological period.

[0109] In a specific example, if Figure 4 As shown, based on the random forest method, the importance score of each classification feature in the time series classification feature images of multiple categories of each phenological period is calculated, and the first classification feature variable set is determined based on the importance score of each classification feature, including:

[0110] Step S111 : Based on the out-of-bag (OOB) error of the random forest method, the importance score of each classification feature in the time series classification feature images of multiple categories in each phenological period is determined.

[0111] It's important to note that when constructing each decision tree in a random forest, a subset of sample data is randomly sampled from the sample dataset to be used in the decision tree construction. The remaining sample data not sampled is referred to as out-of-bag data. The out-of-bag error (OOB) is used to calculate the crop classification error rate of the random forest model based on this out-of-bag data. This error can be used to evaluate the performance of the decision tree.

[0112] In specific implementation, based on the out-of-bag error of the random forest method, the importance score of each classification feature is determined through the following steps:

[0113] 1) For the classification feature X, for the i-th decision tree, select the out-of-bag data corresponding to the decision tree and calculate the first out-of-bag error, recorded as err1 i .

[0114] 2) Randomly add noise to the out-of-bag data corresponding to the decision tree, that is, randomly change the value of the sample data in the classification feature X, and calculate the second out-of-bag error, recorded as err2 i .

[0115] 3) According to the formula:

[0116]

[0117] The importance score I of the classification feature X is calculated. The larger the importance score I, the greater the impact of the classification feature X on the prediction results of crop classification and the higher its importance.

[0118] Where N is the total number of decision trees in the random forest; i is the i-th decision tree; err1 i Indicates the first out-of-bag error of the i-th decision tree; err2 i represents the second out-of-bag error of the i-th decision tree.

[0119] In practical applications, for a classification feature X, the importance score I of the classification feature X is calculated K times according to the above steps. The average of these K importance scores I is used as the final importance score of the classification feature X, where K is a positive integer. This can avoid the randomness of the random forest method and improve the accuracy of feature screening.

[0120] Step S112: sort the importance scores in descending order, and remove the classification features corresponding to the M importance scores with the lowest order, to obtain a new classification feature variable subset; wherein M is an integer greater than 1.

[0121] By removing the classification features corresponding to the M importance scores that are ranked lower, the dimension of the classification features can be reduced.

[0122] Step S113: Determine the classification accuracy of the target overwintering crops based on the new subset of classification feature variables.

[0123] Step S114, iteratively execute steps S111 to S113 until the classification accuracy of the target overwintering crops converges, and use the classification feature variable subset corresponding to the time when the classification accuracy of the target overwintering crops converges as the first classification feature variable set.

[0124] According to the above steps, all temporal classification features of each phenological period are screened to obtain the classification feature variable set of each phenological period.

[0125] In the specific implementation, for the 63 classification feature variables corresponding to the 63 time series feature images, after the above-mentioned feature screening steps, the classification feature variable set of each phenological period is obtained. Among them, the classification feature variable set used to identify the sowing period of the target overwintering crops includes the following classification feature variables: NDTI, BSI, NDPI, WPDI 3-1 、WPDI 2-1 DOY max , Slope; the set of categorical feature variables used to identify the early stage of overwintering includes the following categorical feature variables: BSI, LSWI, WPDI 3-1 、WPDI 2-1 DOY max , Slope; The set of categorical feature variables used to identify the greening period includes the following categorical feature variables: RENDVI, NIR, REP, WPDI 3-1 、WPDI 2-1 DOY max 、Slope。

[0126] Among them, WPDI 2-1 、WPDI 3-1The scores were ranked high in all the importance scorings, indicating that the crop phenology difference index proposed in the embodiment of the present application has an important influence on the accurate identification of overwintering crops. The Normalized Difference Tillage Index (NDTI) can reflect the residue coverage of farmland, and is used to reflect the planting area of the crops in the previous planting season of the cultivated land area. The Bare Soil Index (BSI) can highlight the characteristics of bare land and residential land, and has a higher importance score in the early winter and greening period, and is used to distinguish overwintering crops from bare land. The Normalized Diferential Senescent Vegetation Index (NDSVI) and the Land Surface Water Index (LSWI) can reflect the water content information of crops. In addition, LSWI can also reflect the information of soil water content and surface water, and is used to reflect the growth characteristics of winter wheat in the greening period. The Red Edge Normalized Difference Vegetation Index (RENDVI) and Red Edge Position (REP) are calculated based on the red edge band. RENDVI reflects the chlorophyll content of the winter wheat canopy, while REP reflects the nitrogen content of the winter wheat. RENDVI and REP have high importance scores during the early overwintering and greening phases, and are used to effectively identify overwintering crops during these two phenological periods.

[0127] Step S103: Determine a set of enhanced phenological characteristic indices for each phenological period based on the set of classification characteristic variables for each phenological period.

[0128] It can be understood that each classification characteristic variable in the classification characteristic variable set for each phenological period is obtained by screening based on the time-series classification characteristic image. The classification characteristic variable has a corresponding relationship with the time-series classification characteristic image. In other words, each classification characteristic variable corresponds to a set of classification characteristic images arranged in time. In practical applications, the time-series classification characteristic images corresponding to the classification characteristic variable set can be further screened to obtain an enhanced phenological characteristic index set, or the time-series classification characteristic images corresponding to the classification characteristic variables in the classification characteristic variable set can be synthesized to obtain an enhanced phenological characteristic index set.

[0129] In some embodiments, based on the set of classification feature variables for each phenological period, a set of enhanced phenological feature indices for each phenological period is determined, including: synthesizing the time-series classification feature images corresponding to each classification feature variable in the set of classification feature variables for each phenological period to obtain a set of enhanced phenological feature indices for each phenological period; wherein the classification feature variables in the set of classification feature variables correspond one-to-one to the enhanced phenological feature indices in the set of enhanced phenological feature indices.

[0130] The synthesis process may be maximum value synthesis, minimum value synthesis or median value synthesis.

[0131] In practical applications, different synthesis processing methods can be selected according to the actual content expressed by different classification feature variables. For example, the maximum value synthesis can be performed on the following time series classification feature images: NDPI, LSWI, NDSVI, REP, WPDI 3-1 、WPDI 2-1 DOY max , Slope. The following time series classification feature images can be median-synthesized: NDTI, BSI, RENDVI, NIR. This will yield a set of enhanced phenological characteristic indices for each phenological period. The sum of the enhanced phenological characteristic indices for all phenological periods includes 12 characteristic variables, namely NDPI, LSWI, NDSVI, REP, WPDI 3-1 、WPDI 2-1 DOY max , Slope, NDTI, BSI, RENDVI, NIR.

[0132] Step S104: training a pre-constructed first crop classification model according to the enhanced phenological characteristic index set, so as to classify the target overwintering crops based on the trained first crop classification model.

[0133] In a specific embodiment, according to the enhanced phenological characteristic index set, a characteristic image of any year is selected from time-series images of multiple years as a characteristic band to train a first crop classification model to obtain a trained first crop classification model.

[0134] In another embodiment, according to the enhanced phenological characteristic index set, a first crop classification model for each year is constructed based on characteristic images of multiple years; then, based on the recognition accuracy of the first crop classification model of each year, the crop classification model with the highest accuracy is screened out as the trained first crop classification model.

[0135] In practical applications, the first crop classification model can be constructed based on any one of the methods including support vector machine (SVM), random forest (RF) method, and gradient boosting iterative decision tree (GBDT).

[0136] Among them, the support vector machine (SVM) obtains the optimal hyperplane in the crop classification feature space through statistical learning to determine the maximum interval between different crop classifications.

[0137] According to the formula:

[0138]

[0139] Determine the optimal hyperplane for SVM.

[0140] Where, T is a pre-acquired training sample set. In the embodiment of the present application, each enhanced phenological characteristic index is used as the classification feature of the training sample set; f(C i , T) is to judge whether the training sample belongs to the i-th category C i The probability of f(C j , T) is used to determine whether the training sample belongs to the jth crop category C j probability.

[0141] The random forest algorithm is an ensemble learning algorithm that integrates many decision trees. Each tree uses a guided sampling strategy to create approximately 2 / 3 of the training samples from the original training dataset. A separate decision tree is generated for each training sample, and the remaining approximately 1 / 3 of the training samples are used as out-of-bag data for internal cross-checking. Finally, a vote is conducted based on the classification results of all decision trees, and the classification result with the most votes is selected as the final classification result.

[0142] The GBDT algorithm can be seen as a combination of "gradient boosting" and "decision tree". In essence, it consists of multiple decision trees, each of which is a crop classifier. The classification results of all decision trees are accumulated to obtain the crop classification result. For a complex task, the combined decisions of multiple experts are better than the result of any single one, so this algorithm can be seen as an additive model composed of M decision trees.

[0143] According to the formula:

[0144]

[0145] Calculate crop classification results.

[0146] Where M is the total number of decision trees; s is the pre-acquired training sample; t m is the mth decision tree; w m is the weight of the mth decision tree; α m is the preset parameter of the mth decision tree.

[0147] The enhanced phenological characteristic index set is input into the first crop classification model constructed based on any of the above methods for training, and the trained first crop classification model is obtained for training. Then, the target overwintering crops to be classified are classified based on the trained first crop classification model to obtain the classification results.

[0148] In some embodiments, before training the pre-constructed first crop classification model according to the enhanced phenological characteristic index set, the method further includes: tuning the hyperparameters of the pre-constructed first crop classification model by using a grid search algorithm.

[0149] In practical applications, when constructing a first crop classification model based on SVM to classify remote sensing images, a radial basis function (RBF) is typically selected as the kernel function to achieve a relative balance between time efficiency and accuracy. The cost parameters of the RBF kernel function include c and gamma, where c controls the complexity and versatility of the crop classification model, and gamma controls the range and width of the input space of the crop classification model. For different parameter values, a large c value may lead to overfitting and increase the time cost of the crop classification model, while an inappropriate gamma value may lead to insufficient model accuracy or introduce errors. SVM classification accuracy is sensitive to parameter settings, and parameter optimization is required before classification to achieve the best classification results.

[0150] In the embodiment of the present application, a grid search algorithm is used to optimize the values of c and gamma, and the optimization results are as follows: Figure 5 shown.

[0151] For the first crop classification model built based on the random forest algorithm, the number of randomly selected elements (split) used to split each node and the number of decision trees (ntrees) are two key parameters. In the embodiment of the present application, the grid search algorithm is used to tune split and ntrees. The tuning results are as follows: Figure 6 shown.

[0152] For the first crop classification model built based on the gradient boosting iterative decision tree algorithm, optimizing the number of decision trees (ntrees), learning rate (Lr), and maximum depth of the decision tree (maxDepth) can improve the performance of the model. In the embodiment of the present application, a grid search algorithm is used to tune ntrees, Lr, and maxDepth. The tuning results are shown in Figure 2. Figure 7 shown.

[0153] In this way, by optimizing the key parameters of the first crop classification model, the first crop classification model can achieve relatively good performance. In the embodiment of the present application, the classification accuracy of the crop classification models constructed based on the three algorithms of SVM, RF, and GBDT all reached above 95%.

[0154] In other embodiments, the method further includes: classifying other overwintering crops other than the target overwintering crop based on a pre-constructed second crop classification model to obtain classification results of the other overwintering crops; superimposing the classification results of the target overwintering crop with the classification results of the other overwintering crops to determine the overlapping area between the classification results of the target overwintering crop and the classification results of the other overwintering crops; and removing the overlapping area from the classification results of the target overwintering crop to obtain the final classification result of the target overwintering crop. The final classification result of the target overwintering crop can be represented by a probability graph, such as Figure 8 .

[0155] The second crop classification model can be constructed based on any of the following methods: support vector machine (SVM), random forest (RF), or gradient boosting iterative decision tree (GBDT). For each category of other overwintering crops, a corresponding second crop classification model is constructed. The classification results of all categories of other overwintering crops are then superimposed with the classification results of the target overwintering crop to determine the overlapping areas. These overlapping areas are then removed from the classification results of the target overwintering crop to obtain the classification results of the target overwintering crop.

[0156] In practical applications, the second crop classification model is trained using a second enhanced phenological characteristic index set. The second enhanced phenological characteristic index set can be obtained using the aforementioned method for obtaining the enhanced phenological characteristic index set corresponding to each phenological period of the target overwintering crop. It should be understood that the crop classification features in the second enhanced phenological characteristic index set may be the same as or different from the classification features corresponding to each phenological period of the target overwintering crop.

[0157] For example, for rapeseed, a winter crop grown in the same flowering period as winter wheat, the rapeseed recognition model is trained by obtaining the CFI index and NDYI index of the rapeseed during its flowering period to obtain a second crop recognition model corresponding to the rapeseed.

[0158] In this way, the accuracy of the classification results can be further improved by excluding other overwintering crops with similar phenological characteristics from the classification results of the target overwintering crops.

[0159] Finally, the accuracy of the final classification results of the target winter crops is determined by counting the planting areas of the final classification results of the target winter crops in each planting area and comparing them with the area data of the corresponding crops in the statistical yearbooks released by the official (such as the municipal statistical department). Figure 9 、 Figure 10 、 Figure 11 As shown, the area data obtained by identifying the target overwintering crops using the method provided in the embodiment of the present application is highly consistent with the area data released officially, and R2 is maintained above 0.97, indicating that the method provided in the embodiment of the present application has high accuracy and can meet the needs of practical applications.

[0160] In some optional embodiments, time series images of other years that have not participated in the training are classified based on the trained first crop classification model to achieve temporal migration of the model and evaluate the temporal generalization performance of the first crop classification model.

[0161] In summary, the present application provides an early classification method for overwintering crops based on enhanced phenological characteristics, including: determining multiple categories of time-series classification feature images corresponding to multiple phenological periods in the early growth stage of target overwintering crops based on pre-acquired time-series remote sensing images; performing feature screening on multiple categories of time-series classification feature images to determine a set of classification feature variables for each phenological period; determining a set of enhanced phenological feature indexes for each phenological period based on the set of classification feature variables for each phenological period; training a pre-constructed first crop classification model according to the enhanced phenological feature index, and classifying the target overwintering crops based on the trained first crop classification model. In this way, after acquiring the time-series remote sensing images, multiple categories of time-series classification feature images corresponding to each phenological period are determined according to multiple phenological periods in the early growth stage of the target overwintering crops. The classification feature images for different phenological periods are different, thereby ensuring that the classification features are adapted to the phenological period of the crop to the greatest extent. Among them, multiple categories of time-series classification feature images include at least phenological change feature images, so as to address the problem of weak vegetation characteristics of overwintering crops in the early stage, enhance the vegetation characteristics in subsequent steps, and improve the early recognition accuracy of overwintering crops.

[0162] In this application, characteristic variables with higher contribution are analyzed more accurately according to the phenological characteristics of the early growth stage of crops, and a classification model is constructed based on these characteristic variables to identify crop types, which helps to improve the accuracy and efficiency of classification and also enhances the interpretability of the model.

[0163] Exemplary Systems

[0164] The present application embodiment provides an early classification system for overwintering crops based on enhanced phenological characteristics. Figure 12This is a schematic diagram of the structure of an early classification system for overwintering crops based on enhanced phenological characteristics according to some embodiments of the present application. Figure 12 As shown, the system includes: a first calculation unit 1201, a feature screening unit 1202, a second calculation unit 1203, and a crop classification unit 1204.

[0165] The first calculation unit 1201 is configured to determine, based on pre-acquired time-series remote sensing images, multiple categories of time-series classification feature images corresponding to multiple phenological periods in the early growth stage of the target overwintering crop; wherein the multiple phenological periods in the early growth stage of the target overwintering crop include at least a first phenological period and a second phenological period, the first phenological period represents the sowing period of the target overwintering crop, and the second phenological period represents the early overwintering period of the target overwintering crop or the greening period of the target overwintering crop; the time-series classification feature images include at least phenological change feature images.

[0166] The feature screening unit 1202 is configured to perform feature screening on the multiple categories of time-series classification feature images to determine a set of classification feature variables for each of the phenological periods;

[0167] The second calculation unit 1203 is configured to determine a set of enhanced phenological characteristic indices for each phenological period based on the set of classification characteristic variables for each phenological period;

[0168] The crop classification unit 1204 is configured to train a pre-constructed first crop classification model according to the enhanced phenological characteristic index set, so as to classify the target overwintering crop based on the trained first crop classification model.

[0169] The early classification system for overwintering crops based on enhanced phenological characteristics provided in the embodiments of the present application can implement the processes and steps of any of the above-mentioned early classification methods for overwintering crops based on enhanced phenological characteristics, and achieve the same technical effects, which will not be repeated here.

[0170] The foregoing description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A method for early classification of overwintering crops based on enhanced phenological characteristics, characterized in that: include: Based on pre-acquired time-series remote sensing images, multiple categories of time-series classification feature images corresponding to multiple phenological periods in the early growth stage of target overwintering crops are determined; wherein the multiple phenological periods in the early growth stage of the target overwintering crops include at least a first phenological period and a second phenological period, the first phenological period represents the sowing period of the target overwintering crops, and the second phenological period represents the early overwintering period of the target overwintering crops or the greening period of the target overwintering crops; the multiple categories of time-series classification feature images include at least phenological change feature images; Performing feature screening on the multiple categories of time-series classification feature images to determine a set of classification feature variables for each of the phenological periods; Determining a set of enhanced phenological characteristic indices for each phenological period based on a set of classification characteristic variables for each phenological period; Training a pre-constructed first crop classification model according to the enhanced phenological characteristic index set, classifying the target overwintering crop based on the trained first crop classification model, and obtaining a classification result of the target overwintering crop; The phenological change characteristics include the crop phenological difference index; the crop phenological difference index is calculated according to the following formula: ， Wherein, WPDI represents the crop phenological difference index; time series data representing the normalized phenological index of the first phenological period, Time series data representing the normalized phenological index of the second phenological period; The phenological change characteristics include the crop phenological ratio index; the crop phenological ratio index is calculated according to the following formula: ， Wherein, WPRI represents the crop phenology ratio index; time series data representing the normalized phenological index of the first phenological period, Time series data representing the normalized phenological index of the second phenological period; The phenological change characteristics also include the annual cumulative days corresponding to the maximum value of the normalized phenological index in the early wintering time window; according to the formula: ， Determining the annual accumulated day corresponding to the maximum value of the normalized phenological index within the pre-wintering time window; Where, DOY max The annual cumulative days corresponding to the maximum value of the normalized phenological index in the pre-wintering time window; NDPI pho represents the normalized phenological index; pho_3 represents the sowing period of the target overwintering crop; pho_4 represents the early overwintering period of the target overwintering crop.

2. The method for early classification of overwintering crops based on enhanced phenological characteristics according to claim 1, characterized in that: The feature screening of the multiple categories of time-series classification feature images to determine the classification feature variable set of each phenological period includes: Calculating, based on a random forest method, an importance score of each classification feature in the time-series classification feature images of the multiple categories of each phenological period, and determining a first classification feature variable set based on the importance score of each classification feature; Based on the analysis method of correlation analysis, the first classification characteristic variable set is screened for correlation to determine the classification characteristic variable set of each phenological period.

3. The early classification method for overwintering crops based on enhanced phenological characteristics according to claim 2, characterized in that: The step of calculating the importance score of each classification feature in the time-series classification feature images of the multiple categories in each phenological period based on the random forest method, and determining the first classification feature variable set based on the importance score of each classification feature, includes: Step S111, determining the importance score of each classification feature in the time series classification feature images of the multiple categories in each phenological period based on the out-of-bag error of the random forest method; Step S112: sort the importance scores in descending order, and remove the classification features corresponding to the M importance scores with the lowest order, to obtain a new classification feature variable subset; wherein M is an integer greater than 1; Step S113: determining the classification accuracy of the target overwintering crop based on the new classification feature variable subset; Step S114, iteratively executing steps S111 to S113 until the classification accuracy of the target overwintering crops converges, and using the classification feature variable subset corresponding to the time when the classification accuracy of the target overwintering crops converges as the first classification feature variable set.

4. The method for early classification of overwintering crops based on enhanced phenological characteristics according to claim 1, characterized in that: The step of determining the enhanced phenological characteristic index set for each phenological period based on the classification characteristic variable set for each phenological period includes: The time series classification feature images corresponding to each classification feature variable in the classification feature variable set of each phenological period are synthesized to obtain an enhanced phenological feature index set of each phenological period; wherein the classification feature variables in the classification feature variable set correspond one-to-one to the enhanced phenological feature indices in the enhanced phenological feature index set.

5. The method for early classification of overwintering crops based on enhanced phenological characteristics according to claim 1, characterized in that: Before training the pre-built first crop classification model according to the enhanced phenological characteristic index set, the method further includes: The hyperparameters of the pre-built first crop classification model are tuned by a grid search algorithm.

6. The method for early classification of overwintering crops based on enhanced phenological characteristics according to claim 1, characterized in that: The method further comprises: Based on the pre-built second crop classification model, other overwintering crops other than the target overwintering crop in the same period are classified to obtain classification results of other overwintering crops; Superimposing the classification result of the target overwintering crop with the classification results of the other overwintering crops to determine an overlapping area between the classification result of the target overwintering crop and the classification results of the other overwintering crops; The overlapping area is removed from the classification results of the target overwintering crops.

7. An early classification system for overwintering crops based on enhanced phenological characteristics, characterized in that: include: The first computing unit is configured to determine, based on pre-acquired time-series remote sensing images, a plurality of categories of time-series classification characteristic images corresponding to a plurality of phenological periods in the early growth stage of a target overwintering crop, wherein the plurality of phenological periods in the early growth stage of the target overwintering crop include at least a first phenological period and a second phenological period, the first phenological period representing the sowing period of the target overwintering crop, and the second phenological period representing the early overwintering period of the target overwintering crop or the greening period of the target overwintering crop; the time-series classification characteristic images include at least phenological change characteristic images; a feature screening unit configured to perform feature screening on the time series classification feature images of the multiple categories to determine a set of classification feature variables for each of the phenological periods; a second calculation unit configured to determine a set of enhanced phenological characteristic indices for each phenological period based on a set of classification characteristic variables for each phenological period; a crop classification unit configured to train a pre-constructed first crop classification model according to the enhanced phenological characteristic index set, so as to classify the target overwintering crop based on the trained first crop classification model and obtain a classification result of the target overwintering crop; The phenological change characteristics include the crop phenological difference index; the crop phenological difference index is calculated according to the following formula: ， Wherein, WPDI represents the crop phenological difference index; time series data representing the normalized phenological index of the first phenological period, Time series data representing the normalized phenological index of the second phenological period; The phenological change characteristics include the crop phenological ratio index; the crop phenological ratio index is calculated according to the following formula: ， Wherein, WPRI represents the crop phenology ratio index; time series data representing the normalized phenological index of the first phenological period, Time series data representing the normalized phenological index of the second phenological period; The phenological change characteristics also include the annual cumulative days corresponding to the maximum value of the normalized phenological index in the early wintering time window; according to the formula: ， Determining the annual accumulated day corresponding to the maximum value of the normalized phenological index within the pre-wintering time window; Where, DOY max The annual cumulative days corresponding to the maximum value of the normalized phenological index in the pre-wintering time window; NDPI pho represents the normalized phenological index; pho_3 represents the sowing period of the target overwintering crop; pho_4 represents the early overwintering period of the target overwintering crop.

Images