Collaborative identification method and device for soybeans and peanuts in seasons

Abstract

The invention relates to a collaborative recognition method and device for soybeans and peanuts in seasons, and belongs to the technical field of agricultural remote sensing. According to the method, through ingenious design of a three-dimensional coupling coordination degree and a crop canopy coverage index, a large-scale soybean and peanut sample generation module is constructed, a leguminous crop collaborative identification technology process is established, and a multi-year leguminous crop after-season distribution result is obtained. A soybean and peanut distribution reasoning model based on space-time perception is designed, multi-year stable training samples and feature intervals of soybeans and peanuts are obtained in combination with multi-year post-season distribution results of leguminous crops, and a historical knowledge constrained soybean and peanut intra-season collaborative recognition method is constructed. The method has interpretability and large-scale cross-domain migration capability, and a technical method and device are provided for obtaining large-scale and high-precision soybean and peanut planting distribution information in a high-time-efficiency manner.

Description

Technical Field

[0001] This invention belongs to the field of agricultural remote sensing technology, specifically relating to a method and device for collaborative identification during the soybean and peanut seasons. Background Technology

[0002] Leguminous crops such as soybeans and peanuts are crucial for national food and oil security and the stable supply of edible vegetable oils. Their unique biological nitrogen-fixing capacity reduces reliance on chemical nitrogen fertilizers, improves soil fertility and crop yield and quality, and helps reduce greenhouse gas emissions, playing a vital role in modern sustainable agricultural systems. However, my country's high dependence on imported soybeans poses a serious supply risk and challenges from international market fluctuations. Peanuts, as a high-yield, highly adaptable, and relatively self-sufficient oilseed crop, are a key breakthrough for alleviating supply and demand imbalances and tapping into domestic production potential. Rapidly and accurately obtaining information on the spatiotemporal distribution of soybean and peanut planting is of great significance for maintaining national food security.

[0003] Existing soybean identification algorithms include: Huang et al., who constructed an automatic soybean phenological mapping algorithm (PSCC) combining canopy moisture and chlorophyll changes based on soybean phenological characteristics; Chen et al., who created a comprehensive greenness and moisture content index (GWCCI) by using the product of the vegetation index NDVI and shortwave infrared bands; and Xiao et al., who determined the globally optimal time window and integrated soybean chlorophyll content, canopy moisture content, and canopy greenness characteristics to construct a comprehensive soybean mapping index (SMCI). Currently, peanut remote sensing mapping algorithms are still in the exploratory stage. Existing crop identification algorithms need to determine the optimal identification time window based on prior knowledge, making it difficult to flexibly cope with the spatiotemporal heterogeneity of crop phenology in different regions and years.

[0004] Large-scale, high-precision spatial distribution mapping of soybeans and peanuts faces multiple challenges: (1) Soybeans and peanuts have highly similar spectral characteristics to dryland crops, making it difficult to effectively distinguish legumes and identify soybeans and peanuts in a coordinated manner; (2) The heterogeneity of phenological and spectral characteristics of crops in different regions makes it difficult for existing model methods to achieve high-precision application in the case of few or no samples; (3) Existing soybean and peanut mapping algorithms have low timeliness, and most datasets are post-season results, lacking efficient mid-season identification algorithms and datasets. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings and challenges of the prior art by providing a method and apparatus for mid-season collaborative identification of soybeans and peanuts. This method creatively starts from the crop growth mechanism, explores and quantifies the unique three-dimensional spectral coupling patterns and canopy coverage differences of leguminous crops, proposes a three-dimensional coupling coordination degree and a crop canopy coverage index, designs a high-confidence soybean and peanut detection method, and determines the preferred feature set of soybeans and peanuts. Then, it proposes a positive and negative sample generation module for soybeans and peanuts and a collaborative identification technology process for leguminous crops with cross-year migration capability. Furthermore, it designs a soybean and peanut distribution inference model based on spatiotemporal awareness, and finally achieves large-scale, high-precision mid-season collaborative identification of soybeans and peanuts.

[0006] To achieve the above objectives, the technical solution of the present invention is: a method for synergistic identification during the soybean and peanut seasons, comprising:

[0007] Step S01: Design a three-dimensional coupling compatibility index based on bright-green-wet simultaneous heterogeneity;

[0008] Step S02: Construct a crop canopy coverage index based on time-series analysis of soil background information;

[0009] Step S03: Design an automatic positive sample generation module that integrates three-dimensional coupling coordination degree and canopy coverage;

[0010] Step S04: Create a preferred feature set for soybeans and peanuts;

[0011] Step S05: Design a large-scale high-confidence sample intelligent generation module;

[0012] Step S06: Construct a collaborative identification technology process for legume crops;

[0013] Step S07: Design a spatiotemporal awareness-based soybean and peanut distribution inference model;

[0014] Step S08: Construct a collaborative identification method for soybean and peanut seasons with historical knowledge constraints.

[0015] Furthermore, in step S01, a three-dimensional coupling coordination index based on the simultaneous heterogeneity of brightness-greenness-wetness is constructed to address the differences in the synergistic responses of multidimensional attributes among different crops during their growth period. For soybeans and peanuts, during the peak growth period, due to the flat leaves, high chlorophyll content, and weak transpiration, they exhibit a unique three-dimensional coupling characteristic of "high brightness-high greenness-low humidity". The calculation formula for the three-dimensional coupling coordination coordination index BGW3DCC (Brightness Greenness and Wetness High-High-Low 3D Coupling Coherence, BGW3DCC) based on the simultaneous heterogeneity of brightness-greenness-wetness is as follows:

[0016]

[0017]

[0018] in, , and They represent Brightness, Greenness, and Wetness values ​​at specific points in time. , and These represent the threshold values ​​for the corresponding indices. , and Estimation is based on the density distribution characteristics of field sample data. This indicates that the crop meets the three-dimensional coupling time period of "high brightness-high greenness-low humidity". Indicates the crop growing season. , T represents respectively BGW3DCC T pot Number of observations within a time period.

[0019] Furthermore, in step S02, based on the optimal time window T Key By combining the Dry Bare Soil Index (DBSI), a Crop Canopy Cover Index (CCI) is constructed to effectively distinguish between soybeans and peanuts. The calculation formula is as follows:

[0020]

[0021] in, They represent DBSI values ​​at specific time points; coupling soybean and peanut brightness-greenness-humidity three-dimensionally over a time period T. BGW3DCC The optimal time window T for soybean and peanut identification was determined. Key .

[0022] Furthermore, in step S03, the target area is first preliminarily identified based on the three-dimensional coupling coordination index to screen candidate areas for leguminous crops. Then, a crop canopy coverage index is introduced within the candidate areas to further distinguish between soybeans and peanuts, obtaining high-confidence detection results. The detection rules are as follows:

[0023]

[0024] in, Based on the three-dimensional coupling coordination index With crop canopy coverage index The classification discriminant function outputs values ​​1, 2, and 3, corresponding to soybeans, peanuts, and other crops, respectively; the threshold... , The density distribution characteristics of the field sample data were determined; confidence constraints were applied to randomly generated crop point samples to obtain positive samples of soybeans and peanuts with high confidence.

[0025] Furthermore, in step S04, based on a combination of mechanistic understanding and data-driven approach, an optimized feature set for identifying differences between soybeans and peanuts is constructed. First, heterogeneous features are extracted from optical and microwave remote sensing data, focusing on multidimensional attributes including crop growth characteristics, biochemical and physiological states, and canopy structure. Specifically, based on Sentinel-2 MSI, bands including Blue, Green, Red, red-edge, near-infrared, and short-wave infrared are selected to construct spectral index features reflecting vegetation growth, pigment content, soil exposure, and water status. Simultaneously, Sentinel-1 radar backscattering coefficients are introduced to extract microwave scattering features characterizing crop canopy structure and water content. Based on this, temporal statistical analysis is performed on the heterogeneous features throughout the crop's entire growth period, calculating the feature mean image to mitigate the impact of short-term temporal fluctuations and obtain a temporally stable original feature image set. Further, combined with field sample point data, the Jeffries–Matusita method is used... Distance is used to quantitatively evaluate the feature separability between target crop categories and non-target categories. Feature subsets with optimal discriminative ability for soybeans and peanuts are selected and optimized feature sets for soybeans and peanuts are constructed to provide highly discriminative and low-redundancy feature inputs for subsequent classification.

[0026] Furthermore, in step S05, a sample intelligent generation module combining confidence level and feature distribution dual constraints is constructed to achieve the automatic construction of large-scale, high-confidence soybean and peanut positive and negative samples. Randomly generated crop point samples within the cultivated land mask area are used as a candidate sample set, and the candidate samples are screened and purified through positive sample construction mechanisms and negative sample construction mechanisms, respectively. Specifically, the positive sample construction mechanism achieves the automatic generation of highly reliable positive samples through the synergistic effect of confidence level constraints and spatial consistency constraints; the negative sample construction mechanism achieves the automatic generation of highly pure negative samples through the synergistic effect of feature distribution constraints and exponential interval constraints. Regarding positive sample construction, firstly, based on the high-confidence soybean and peanut detection results obtained in step S03, confidence level constraints are applied to the randomly generated crop point samples to obtain initial candidate positive samples. Based on this, an S×S... The spatial neighborhood window is used to perform consistent convolution filtering on the spatial distribution of candidate positive samples to suppress spatially isolated pixels and locally misclassified samples, thereby obtaining positive soybean and peanut samples with good spatial continuity and high confidence. Regarding negative sample construction, based on the preferred feature sets of soybean and peanut determined in step S04, and combined with real soybean and peanut samples, a probability density distribution model for each feature dimension is constructed, and feature distribution constraints are applied to random crop point samples accordingly. When a candidate sample simultaneously meets the distribution discrimination conditions of "non-soybean" and "non-peanut" in all feature dimensions, it is identified as a potential negative sample. Simultaneously, by checking whether the three-dimensional coupling coordination index of the potential negative sample falls within the feature interval of non-soybean and non-peanut samples, a secondary constraint is applied to the negative sample to further improve its purity and reliability.

[0027] Furthermore, in step S06, a collaborative identification technology process for legume crops is constructed to obtain the post-season spatial distribution results of legume crops on a multi-year scale. To improve the model's generalization ability under different year conditions, an interannual migration enhancement strategy is introduced in the model training and application stages, and a system design is carried out from three levels: feature processing, model construction, and training strategy. At the feature processing level, phenological alignment and interannual normalization are performed on the multi-source temporal features of soybeans and peanuts from multiple source years to weaken the feature distribution drift caused by factors including differences in phenological processes, changes in climate conditions, and inconsistencies in remote sensing observation conditions, thereby enhancing the statistical consistency of feature expressions in different years. At the model construction level, the year is used as a variable of observation conditions. The model training process incorporates quantitative methods. By imposing year-insensitive constraints, the model's dependence on year-specific information is limited, guiding it to focus more on the stable growth characteristics and canopy structure differences of the crop itself, thereby learning crop discrimination representations with year invariance. At the model training level, the model is jointly trained based on high-confidence positive and negative sample data from multiple source years, allowing the model to be constrained and optimized simultaneously under multi-year conditions, resulting in a crop identification model with stable discrimination capabilities for different years. Finally, the crop identification model with cross-year transferability is directly applied to remote sensing image data of the target year, enabling the identification of the post-season spatial distribution of leguminous crops on a multi-year scale without the need for target year samples to participate in training.

[0028] Furthermore, in step S07, to automatically construct reliable crop season identification training samples from stable planting areas at a multi-year scale, a spatiotemporal awareness-based soybean and peanut distribution inference model is constructed. This inference model employs a difference-in-differences (DID) model. Based on the multi-year legume distribution results obtained through the legume collaborative identification technology process constructed in step S06, the model identifies high-confidence regions that maintain stable planting characteristics in both time and space through a combination of temporal inference and spatial constraints. The multi-year post-season spatial distribution results of leguminous crops obtained from the constructed leguminous crop collaborative identification technology process are used as input. Utilizing multi-year temporal information and spatial correlation, a difference-in-differences model is employed to mitigate non-crop disturbances, including interannual climate fluctuations, sensor differences, and data noise. The inference model, based on the parallel trend assumption, uses the temporal superposition of multi-year classification results to select pixels with similar spectral characteristics between T1 and T2 that have not undergone land cover change for many years, thus constructing a control group for the difference-in-differences model. Subsequently, the differences in spectral and radar backscattering characteristics between the target pixels and the control group are calculated in the two time periods, and the true crop change signal is extracted through two difference operations. The calculation formula is as follows:

[0029]

[0030]

[0031]

[0032] Where x represents the feature value of the target pixel; μ and This represents the mean and covariance matrix of the control group's eigenvalues. This represents the Mahalanobis distance between the target pixel x and the control group H; , , and These represent the feature values ​​of the target pixel and the control group for years T1 and T2, respectively. and This represents the change before and after the change; the change before and after the change is subtracted a second time, and the absolute value is taken to obtain the change DD:

[0033]

[0034] The above formula characterizes the difference in the degree of change of the spectral and radar backscattering features between the target pixel and the control group from T1 to T2. Its purpose is to perform a second difference operation on the interannual variation to offset the influence of regional or systematic interannual variation. Based on the stable planting area over many years, a spatial neighborhood convolution strategy is further introduced to screen and constrain the spatial continuous area, and automatically generate a multi-year continuous training sample set of soybeans and peanuts to support the training and optimization of the crop season identification model.

[0035] Furthermore, in step S08, firstly, the Brightness-Greenness-Humidity Difference Index (BGWDI) is constructed to characterize the early spectral response of crops, and its calculation formula is as follows:

[0036]

[0037] in, , and They represent Brightness, Greenness, and Wetness values ​​at specific points in time. The moment when the first difference of the Greenness index becomes negative is defined as the moment when the first difference of the Greenness index becomes negative. The early identification window is formed by taking M days before and N days after that moment.

[0038] Secondly, based on the analysis of stable soybean and peanut samples over many years, its Due to historical trajectories, soybeans and peanuts exhibit differences in canopy density in their early stages. A canopy density index (CDI) is constructed to characterize these differences in crop canopy structure and distinguish between soybeans and peanuts in their early stages. The calculation formula is as follows:

[0039]

[0040] in, T represents peak Number of observations within the time period;

[0041] Then, in the BGWDI–CDI feature domain, each cell is represented as a two-dimensional feature vector. Based on years of stable training samples, the distribution characteristics of the crop in the feature space were analyzed. For different crop types c, the stable value ranges over many years were extracted in the BGWDI and CDI dimensions, and the corresponding stable feature spaces were constructed accordingly. Defined as:

[0042]

[0043] in, , and , These represent the lower and upper bounds of the multi-year stable values ​​of soybeans / peanuts in the BGWDI and CDI dimensions, respectively. The boundary parameters are determined by statistical analysis of the density distribution of field samples in the feature space.

[0044] Finally, based on a stable feature space, a rule-based collaborative identification method for mid-season soybeans and peanuts is constructed; when the feature vector x corresponding to a pixel falls into the stable feature space of a certain crop... If x is not in the stable feature space of soybean / peanut, it is classified as the corresponding crop; if x is not in the stable feature space of soybean / peanut, it is classified as another crop; given that the stable feature spaces of soybean and peanut may overlap in local areas, a priority discrimination rule based on the center distance of the feature spaces is further introduced; the center position of soybean / peanut in the BGWDI–CDI feature space is:

[0045]

[0046] When a pixel feature vector x falls within the overlapping region of both the soybean and peanut feature spaces, the Euclidean distance between that pixel and the soybean and peanut feature spaces is calculated separately, and the minimum distance is taken as the discrimination type.

[0047]

[0048] By constructing feature space and establishing discrimination rules, we can achieve collaborative identification of mid-season soybeans and peanuts and reduce confusion between the two crops due to feature similarity during their critical growth periods.

[0049] This invention also provides a soybean-peanut mid-season collaborative identification device, comprising: an optical radar time-series data processing module, a three-dimensional coupling coordination degree index construction module, a multi-year continuous post-season crop mapping module, a soybean-peanut mid-season discrimination model construction module, and a mid-season collaborative identification result output module; wherein,

[0050] The optical radar time-series data processing module is used to acquire and preprocess optical and radar remote sensing time-series data; on the remote sensing cloud platform, Sentinel-1 SAR and Sentinel-2 MSI image data that meet the research time period are selected, and image mosaicking, speckle filtering, cloud removal, cropping, interpolation, and temporal smoothing are performed respectively to generate optical and radar time-series image datasets with uniform temporal resolution; based on the processed Sentinel-2 MSI time-series images, a multi-dimensional spectral index time-series dataset including vegetation index, soil and moisture index, and pigment and browning index is further constructed for subsequent crop feature analysis and identification.

[0051] The 3D Coupling Coordination Index Construction Module is used to construct a 3D coupling coordination index based on the simultaneous heterogeneity of brightness, greenness, and humidity according to the remote sensing time series characteristics of crops throughout their entire growth period, in order to characterize the synergistic response relationship of brightness, greenness, and humidity characteristics during the transition process of growth stages; the multi-year continuous post-season mapping module for crops identifies stable 3D coupling characteristics formed by crops in key growth stages by analyzing the time series changes of the 3D coupling coordination index, and automatically determines the optimal time window, thereby improving the stability and reliability of crop type identification;

[0052] The multi-year continuous post-season mapping module for crops is used to construct a collaborative crop identification process with cross-year migration capabilities to obtain multi-year crop post-season spatial distribution results. This module employs an inter-annual migration enhancement strategy to perform phenological alignment and inter-annual normalization on the temporal features of multiple source years, and applies year-insensitive constraints during model training to improve the model's cross-year generalization ability. The module directly applies the crop identification model trained over multiple years to remote sensing image data of the target year, achieving continuous post-season mapping of crops without the need for samples from the target year.

[0053] The soybean-peanut mid-season discrimination model construction module is used to distinguish between soybeans and peanuts in the early stages. Based on years of stable training samples, it analyzes the historical evolution of crop multidimensional characteristics, extracts stable temporal features of key growth stages, and constructs a feature space and discrimination rules for mid-season collaborative identification. This module constructs the Brightness-Greenness-Humidity Difference Index (BGWDI) and the Canopy Density Index (CDI) to characterize the stable differences of crops in spectral response and canopy structure, and automatically determines stable intervals in the BGWDI-CDI feature space. Combined with rule-based discrimination and overlap distance disambiguation strategies, it achieves reliable differentiation between soybeans and peanuts in the early stages and reduces the risk of confusion during key growth periods.

[0054] The mid-season collaborative identification result output module is used to integrate and output the mid-season collaborative identification results of soybeans and peanuts. It integrates high-confidence identification results obtained based on three-dimensional coupling coordination degree and crop canopy coverage index, constructs an optimized feature set and automatically generates a large-scale positive and negative sample, forming a collaborative identification model with cross-year transfer capability and outputting multi-year post-season distribution results. The multi-year continuous post-season mapping module further combines multi-year distribution results with crop spatiotemporal reasoning mechanism to dynamically obtain the latest year's training samples and extract multi-year stable feature space, realizing the integration and highly reliable output of early soybean and peanut identification results.

[0055] This invention moves beyond the approach of single-crop identification, comprehensively considering the high coupling of soybean and peanut brightness-greenness-moisture dimensions and multi-dimensional attributes such as canopy coverage, and combining knowledge of historical characteristic trajectories of soybeans and peanuts to propose a mid-season collaborative identification method and device for soybeans and peanuts. This invention focuses on the multi-dimensional attributes of soybeans, including brightness, greenness, moisture, and canopy coverage, constructing a three-dimensional coupling coordination degree and a crop canopy coverage index. It then sequentially designs a high-confidence soybean and peanut detection method, a preferred feature set for soybeans and peanuts, and a large-scale high-confidence positive and negative sample intelligent generation module for soybeans and peanuts. Furthermore, it constructs a collaborative identification technology process for legumes with cross-year migration capabilities, and designs a spatiotemporal awareness-based soybean and peanut distribution inference model, ultimately achieving large-scale, high-precision mid-season collaborative identification of soybeans and peanuts.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] (1) Pioneering the “BGW 3D Coupling Coordination Index (BGW3DCC)”: Based on a systematic analysis of the temporal characteristics of tassel transformation (Brightness, Greenness, Wetness) of various crops, the “high brightness-high greenness-low humidity” three-dimensional coupling pattern shared by soybeans and peanuts during their peak growth period was discovered and quantified for the first time. The BGW3DCC index was creatively proposed. This index effectively captures the unique three-dimensional spectral coupling characteristics of legumes through coordination degree calculation, providing a key basis for the accurate identification of legumes.

[0058] (2) Creating a high-confidence, large-scale sample generation strategy: Based on the systematic analysis of multi-dimensional crop features, this invention designs a three-dimensional coupling coordination index, a crop canopy coverage index, and a set of preferred features for soybeans and peanuts. It constructs a sample intelligent generation module that combines confidence and feature distribution constraints, and develops an efficient large-scale sample generation strategy. This strategy solves the problems of sample scarcity and quality bottlenecks, laying the foundation for integrating knowledge and data-driven algorithms.

[0059] (3) Design of soybean and peanut distribution inference model: The method designed in this invention is ingenious. By constructing a soybean and peanut distribution inference model based on spatiotemporal awareness, the interference of interannual fluctuations is eliminated, thereby realizing large-scale cross-domain and cross-year high-precision automatic crop mapping. It solves the technical bottleneck of low model robustness in cross-domain and cross-year automatic migration applications and provides technical support for the automatic business operation of crop identification algorithms.

[0060] (4) Constructing a high-precision mid-season collaborative identification scheme for soybeans and peanuts: Based on the systematic analysis of the historical trajectory of crop multi-dimensional features, the Brightness-Greenness-Humidity Difference (BGWDI) and Canopy Density Index (CDI) were designed. Soybean-peanut discrimination rules were established based on the BGWDI-CDI feature space. Priority discrimination rules based on the center distance of the feature space were introduced to disambiguate overlapping areas, thus constructing a mid-season collaborative identification method for soybeans and peanuts constrained by historical knowledge. This provides a scientific solution for mid-season collaborative identification of soybeans and peanuts. This scheme significantly improves the accuracy and efficiency of early collaborative identification of soybeans and peanuts and provides new research ideas and technical support for early collaborative identification of more specialty crops. Attached Figure Description

[0061] Figure 1 This is a schematic diagram illustrating the design and implementation process of an embodiment of the present invention;

[0062] Figure 2 This is a schematic diagram of the device structure according to an embodiment of the present invention;

[0063] Figure 3 This is a spatial distribution map of soybeans and peanuts during the soybean and peanut season in a certain county of a certain city in a certain province, according to an embodiment of the present invention. Detailed Implementation

[0064] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0065] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0066] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0067] This invention provides a construction process for a soybean-peanut mid-season collaborative identification method and device, and based on this, provides a soybean-peanut mid-season mapping method based on historical knowledge constraints, including the following steps: Figure 1 As shown:

[0068] Step S01: Design a three-dimensional coupling compatibility index based on bright-green-wet simultaneous heterogeneity;

[0069] Step S02: Construct a crop canopy coverage index based on time-series analysis of soil background information;

[0070] Step S03: Design an automatic positive sample generation module that integrates three-dimensional coupling coordination degree and canopy coverage;

[0071] Step S04: Create a preferred feature set for soybeans and peanuts;

[0072] Step S05: Design a large-scale high-confidence sample intelligent generation module;

[0073] Step S06: Construct a collaborative identification technology process for legume crops;

[0074] Step S07: Design a spatiotemporal awareness-based soybean and peanut distribution inference model;

[0075] Step S08: Construct a collaborative identification method for soybean and peanut seasons with historical knowledge constraints.

[0076] Specifically, step S01: Design the three-dimensional coupling coordination index of bright-green-wet;

[0077] To address the differences in the synergistic responses of various multidimensional attributes across different crops during their growth stages, a three-dimensional coupling coordination index based on the simultaneous heterogeneity of brightness, greenness, and wetness was constructed. For soybeans and peanuts, during their peak growth period, due to their flat leaves, high chlorophyll content, and weak transpiration, they exhibit a unique three-dimensional coupling characteristic of "high brightness – high greenness – low humidity." The three-dimensional coupling coordination index (Brightness, Greenness, and Wetness High-High-Low 3D Coupling Coherence, BGW3DCC) was designed, and its calculation formula is as follows:

[0078]

[0079]

[0080] in, , and They represent Brightness, Greenness, and Wetness values ​​at specific points in time. , and These represent the threshold values ​​for the corresponding indices. This indicates that the crop meets the three-dimensional coupling time period of "high brightness-high greenness-low humidity". Indicates the crop growing season. , T represents respectively BGW3DCC T pot Number of observations within the time period. Threshold. , and It can be estimated based on the density distribution characteristics of the field sample data, and the suggested values ​​are 0.8±0.01, 0.08±0.005 and -0.03±0.001.

[0081] Step S02: Construct a crop canopy coverage index based on time-series analysis of soil background information;

[0082] Based on the optimal time window (T) Key Combined with the Dry Bare Soil Index (DBSI), a Crop Canopy Cover Index (CCI) is constructed to effectively distinguish between soybeans and peanuts. The calculation formula is as follows:

[0083]

[0084] in, They represent DBSI values ​​at specific time points; the three-dimensional coupling of lightness-greenness-humidity in soybeans and peanuts over a specific time period (T). BGW3DCC The optimal time window (T) for soybean and peanut identification was determined. Key ).

[0085] Step S03: First, the target area is initially identified based on the three-dimensional coupling coordination index to screen candidate areas for leguminous crops. Then, a crop canopy coverage index is introduced within these candidate areas to further distinguish between soybeans and peanuts, thereby obtaining high-confidence detection results. The detection rules are as follows:

[0086]

[0087] in, This is a classification discriminant function based on the three-dimensional coupling coordination index and the crop canopy coverage index, with output values ​​of 1, 2, and 3 corresponding to soybeans, peanuts, and other crops, respectively; the threshold... , The density distribution characteristics of field sample data are used to determine the data. Confidence constraints are applied to randomly generated crop point samples to obtain positive samples of soybeans and peanuts with high confidence levels. , The suggested values ​​are 0.25±0.01 and 0.2+0.01, respectively.

[0088] Step S04: Create a preferred feature set for soybeans and peanuts;

[0089] Based on a combination of mechanistic understanding and data-driven approach, an optimized feature set for differential identification of soybeans and peanuts was constructed. First, heterogeneous features were extracted from optical and microwave remote sensing data, focusing on multidimensional attributes such as crop growth characteristics, biochemical and physiological states, and canopy structure. Specifically, based on Sentinel-2 MSI, spectral index features reflecting vegetation growth, pigment content, soil exposure, and moisture status were constructed by selecting Blue, Green, Red, red-edge, near-infrared, and short-wave infrared bands. Simultaneously, the Sentinel-1 radar backscattering coefficient was introduced to extract microwave scattering features characterizing crop canopy structure and water content. Based on this, temporal statistical analysis was performed on these heterogeneous features throughout the entire crop growth period, calculating the feature mean image to mitigate the impact of short-term temporal fluctuations and obtain a temporally stable original feature image set. Further combining field sample data, the Jeffries–Matusita distance was used to quantitatively evaluate the feature separability between the target crop category and non-target categories. Feature subsets with optimal discriminative ability for soybeans and peanuts were selected respectively, and optimized feature sets for soybeans and peanuts were constructed to provide highly discriminative and low-redundancy feature inputs for subsequent classification.

[0090] Step S05: Design a large-scale high-confidence sample intelligent generation module;

[0091] This step constructs a sample intelligent generation module combining confidence level and feature distribution dual constraints to achieve the automatic construction of large-scale, high-confidence positive and negative samples for soybeans and peanuts. The method uses randomly generated crop point samples within the cultivated land mask area as a candidate sample set, and filters and purifies these candidate samples through positive and negative sample construction mechanisms, respectively. Specifically, the positive sample construction mechanism achieves the automatic generation of highly reliable positive samples through the synergistic effect of confidence level constraints and spatial consistency constraints; the negative sample construction mechanism achieves the automatic generation of highly pure negative samples through the synergistic effect of feature distribution constraints and exponential interval constraints. Regarding positive sample construction, firstly, based on the high-confidence soybean and peanut detection results obtained in step S03, confidence constraints are applied to the randomly generated crop point samples to obtain initial candidate positive samples. Based on this, an S×S spatial neighborhood window is introduced to perform consistent convolution filtering on the spatial distribution of the candidate positive samples to suppress spatially isolated pixels and locally misjudged samples, thereby obtaining soybean and peanut positive samples with good spatial continuity and high confidence. Regarding negative sample construction, based on the preferred feature sets of soybeans and peanuts determined in step S04, and combined with real soybean and peanut samples, a probability density distribution model for each feature dimension is constructed, and feature distribution constraints are applied to random crop point samples for screening. When a candidate sample simultaneously meets the distribution discrimination conditions of "non-soybean" and "non-peanut" across all feature dimensions, it is identified as a potential negative sample. Simultaneously, by verifying whether the three-dimensional coupling coordination index of the potential negative sample falls within the feature interval of non-soybean and non-peanut samples, a secondary constraint is applied to the negative sample to further improve its purity and reliability. S is recommended to be 3 or 7.

[0092] Step S06: Construct a collaborative identification technology process for legume crops;

[0093] This invention constructs a collaborative identification technology process for legume crops to obtain post-season spatial distribution results of legume crops on a multi-year scale. To improve the model's generalization ability under different year conditions, this invention introduces an interannual transfer enhancement strategy in the model training and application stages, and designs a system from three levels: feature processing, model construction, and training strategy. At the feature processing level, phenological alignment and interannual normalization are performed on multi-source temporal features of soybeans and peanuts from multiple source years to reduce feature distribution drift caused by factors such as differences in phenological processes, changes in climate conditions, and inconsistencies in remote sensing observation conditions, thereby enhancing the statistical consistency of feature expressions across different years. At the model construction level, the year is introduced as an observation condition variable into the model training process. By imposing year-insensitive constraints, the model's dependence on year-difference information is limited, guiding the model to focus more on the stable growth characteristics and canopy structure differences of the crop itself, thus learning crop discrimination representations with year invariance. At the model training level, the model is jointly trained based on high-confidence positive and negative sample data from multiple source years, allowing the model to be constrained and optimized simultaneously under multi-year conditions, thereby obtaining a crop identification model with stable discrimination ability across different years. Finally, the crop identification model with cross-year migration capability is directly applied to remote sensing image data of the target year, enabling the identification of the post-season spatial distribution of leguminous crops on a multi-year scale without the need for samples from the target year to participate in training.

[0094] Step S07: Design a spatiotemporal awareness-based soybean and peanut distribution inference model;

[0095] To automatically construct reliable intra-season crop identification training samples from stable multi-year planting areas, a spatiotemporal perception-based soybean and peanut distribution inference model is developed. This model employs a difference-in-differences (DID) model. Based on the multi-year post-season spatial distribution results of legumes obtained in step S06, the model identifies high-confidence areas maintaining stable planting characteristics in both time and space through a combination of temporal inference and spatial constraints. Using the multi-year post-season spatial distribution results of legumes obtained in step S06 as input, the model leverages multi-year temporal information and spatial correlation to mitigate non-crop disturbances caused by interannual climate fluctuations, sensor differences, and data noise through the DID model. Based on the parallel trend assumption, the model uses the temporal superposition of multi-year classification results to select pixels with similar spectral characteristics from year T1 to year T2 that have not experienced land cover change for many years, thus constructing a control group for the DID model. Subsequently, the differences in spectral and radar backscattering characteristics between the target pixels and the control group are calculated in the two time periods, and the true crop change signal is extracted through two difference operations. The calculation formula is as follows:

[0096] ,

[0097] ,

[0098] ,

[0099] Where x represents the feature value of the target pixel; μ and This represents the mean and covariance matrix of the control group's eigenvalues. This represents the Mahalanobis distance between the target pixel x and the control group H. , , and These represent the feature values ​​of the target pixel and the control group for years T1 and T2, respectively. and This is the change before and after the change. Then, we perform a second difference on the change before and after the change and take the absolute value to obtain the change DD.

[0100] .

[0101] This formula characterizes the difference in the degree of change in spectral and radar backscattering features between the target pixel and the control group from year T1 to T2. Its purpose is to perform a second difference operation on interannual variations to offset the influence of regional or systematic interannual variations. Based on the aforementioned multi-year stable planting area, a spatial neighborhood convolution strategy is further introduced to filter and constrain spatially continuous regions, automatically generating multi-year continuous training sample sets of soybeans and peanuts to support the training and optimization of the crop season identification model.

[0102] Step S08: Construct a collaborative identification method for soybeans and peanuts during the season with historical knowledge constraints.

[0103] First, the Brightness-Greenness-Humidity Difference Index (BGWDI) is constructed to characterize the early spectral response of crops. Its calculation formula is as follows:

[0104] ,

[0105] in, The early identification window is defined as the moment when the first difference of the Greenness index becomes negative, taking M days before and N days after that moment. M and N are recommended to be 10~20 and 30~35, respectively.

[0106] Secondly, based on the analysis of stable soybean and peanut samples over many years, its Historical trajectories show differences in canopy density between soybeans and peanuts in their early stages. A canopy density index (CDI) is constructed to characterize these differences in crop canopy structure and distinguish between soybeans and peanuts in their early stages. The calculation formula is as follows:

[0107] .

[0108] in, T represents peak Number of observations within a time period.

[0109] Then, in the BGWDI–CDI feature domain, each cell is represented as a two-dimensional feature vector. The distribution characteristics of crops in this feature space were analyzed based on stable training samples over many years. For different crop types c, their stable value ranges over many years were extracted in the BGWDI and CDI dimensions, and a corresponding stable feature space was constructed accordingly. Defined as:

[0110] ,

[0111] in, , and , These represent the lower and upper bounds of the multi-year stable values ​​of soybeans / peanuts in the BGWDI and CDI dimensions, respectively. The boundary parameters are determined by statistical analysis of the density distribution of field samples in the feature space.

[0112] Finally, based on the aforementioned stable feature space, a rule-based collaborative identification method for mid-season soybeans and peanuts is constructed. This method works when the feature vector x corresponding to a pixel falls into the stable feature space of a particular crop. If x is not in the stable feature space of soybean / peanut, it is classified as the corresponding crop; if x is not in the stable feature space of soybean / peanut, it is classified as another crop. Given that the stable feature spaces of soybean and peanut may overlap in local regions, a priority discrimination rule based on the center distance of the feature spaces is further introduced. The center position of soybean / peanut in the BGWDI–CDI feature space is...

[0113] .

[0114] When a pixel feature vector x falls within the overlapping region of both the soybean and peanut feature spaces, the Euclidean distance between that pixel and the feature spaces of both soybeans and peanuts is calculated, and the minimum distance is taken as the discrimination type. That is...

[0115] .

[0116] Through the aforementioned feature space construction and discrimination rules, the collaborative identification of mid-season soybeans and peanuts was achieved, and the confusion caused by feature similarity between the two crops during the critical growth period was effectively reduced.

[0117] like Figure 2As shown, the present invention also provides a cooperative identification device during the soybean and peanut season, which is used to perform the monitoring method described in Embodiment 1 above: the device includes:

[0118] Optical Radar Time-Series Data Processing Module: This module acquires and preprocesses optical and radar remote sensing time-series data. Sentinel-1 SAR and Sentinel-2 MSI image data meeting the criteria are selected on the remote sensing cloud platform according to the research time period. Image mosaicking, speckle filtering, cloud removal, cropping, interpolation, and temporal smoothing are then performed to generate optical and radar time-series image datasets with uniform temporal resolution. Based on the processed Sentinel-2 MSI time-series images, multi-dimensional spectral index time-series datasets, including vegetation indices, soil and moisture indices, and pigment and browning indices, are further constructed for subsequent crop feature analysis and identification.

[0119] The 3D Coupling Coordination Index Construction Module is used to construct a 3D coupling coordination index based on the simultaneous heterogeneity of brightness, greenness, and humidity according to remote sensing temporal characteristics throughout the entire crop growth period. This index characterizes the synergistic response relationship between brightness, greenness, and humidity characteristics during the transition between growth stages. By analyzing the time-series changes of the 3D coupling coordination index, this module identifies stable 3D coupling characteristics formed by crops during key growth stages and automatically determines the optimal time window, thereby improving the stability and reliability of crop type identification.

[0120] The multi-year continuous post-season mapping module for crops is used to construct a collaborative crop identification process with cross-year migration capabilities to obtain multi-year post-season spatial distribution results. This module employs an inter-year migration enhancement strategy to perform phenological alignment and inter-year normalization on temporal features from multiple sources, and applies year-insensitive constraints during model training to improve the model's cross-year generalization ability. The module directly applies the crop identification model trained on multiple years to remote sensing image data of the target year, achieving continuous post-season mapping of crops without the need for samples from the target year.

[0121] The soybean-peanut mid-season discrimination model construction module is used for early-stage differentiation between soybeans and peanuts. Based on years of stable training samples, it analyzes the historical evolution of multidimensional crop characteristics, extracts stable temporal features at key growth stages, and constructs a feature space and discrimination rules for mid-season collaborative identification. This module characterizes stable differences in spectral response and canopy structure by constructing the Brightness-Greenness-Humidity Difference Index (BGWDI) and Canopy Density Index (CDI), automatically determining stable intervals in the BGWDI–CDI feature space. Combined with rule-based discrimination and overlap region distance disambiguation strategies, it achieves reliable differentiation between early-stage soybeans and peanuts, reducing the risk of confusion during key growth periods.

[0122] The mid-season collaborative identification result output module integrates and outputs the results of mid-season collaborative identification of soybeans and peanuts. It fuses high-confidence identification results obtained based on three-dimensional coupling coordination degree and crop canopy coverage index, constructs an optimized feature set, and automatically generates a large-scale positive and negative sample set. This forms a collaborative identification model with cross-year transferability and outputs post-season distribution results at a multi-year scale. This module further combines multi-year distribution results with crop spatiotemporal inference mechanisms to dynamically acquire the latest year's training samples and extract multi-year stable feature spaces, achieving the integration and highly reliable output of early soybean and peanut identification results.

[0123] Each of the above modules can be configured as computer program instructions stored in a memory, which, when executed by a processor, implement the aforementioned functions. The apparatus may further include a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the method described in Embodiment 1.

[0124] To verify the effectiveness of this invention, a county in a city of a province was selected as the study area, and a national standard administrative division vector map was used as the base map. Following the above method steps, a spatial distribution map of the soybean and peanut seasons in the study area was created using the monitoring device provided by this invention (e.g., ...). Figure 3 (As shown).

[0125] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for co-identification during the soybean and peanut seasons, characterized in that, include: Step S01: Design a three-dimensional coupling compatibility index based on bright-green-wet simultaneous heterogeneity; Step S02: Construct a crop canopy coverage index based on time-series analysis of soil background information; Step S03: Design an automatic positive sample generation module that integrates three-dimensional coupling coordination degree and canopy coverage; Step S04: Create a preferred feature set for soybeans and peanuts; Step S05: Design a large-scale high-confidence sample intelligent generation module; Step S06: Construct a collaborative identification technology process for legume crops; Step S07: Design a spatiotemporal awareness-based soybean and peanut distribution inference model; Step S08: Construct a collaborative identification method for soybean and peanut seasons with historical knowledge constraints.

2. The method for co-identification of soybeans and peanuts during the soybean-peanut season according to claim 1, characterized in that, In step S01, a three-dimensional coupling coordination index based on the simultaneous heterogeneity of brightness-greenness-moisture is constructed to address the differences in the synergistic responses of multidimensional attributes among different crops during their growth period. For soybeans and peanuts, during the peak growth period, due to the flat leaves, high chlorophyll content, and weak transpiration, they exhibit a unique three-dimensional coupling characteristic of "high brightness-high greenness-low humidity". The calculation formula for the three-dimensional coupling coordination index BGW3DCC based on the simultaneous heterogeneity of brightness-greenness-moisture is as follows: in, , and They represent Brightness, Greenness, and Wetness values ​​at specific points in time. , and These represent the threshold values ​​for the corresponding indices. , and Estimation is based on the density distribution characteristics of field sample data. This indicates that the crop meets the three-dimensional coupling time period of "high brightness-high greenness-low humidity". Indicates the crop growing season. , T represents respectively BGW3DCC T pot Number of observations within a time period.

3. The method for co-identification of soybeans and peanuts during the soybean-peanut season according to claim 1, characterized in that, In step S02, based on the optimal time window T Key By combining the Dry Bare Soil Index (DBSI), a Crop Canopy Cover Index (CCI) is constructed to effectively distinguish between soybeans and peanuts. The calculation formula is as follows: in, They represent DBSI values ​​at specific time points; coupling soybean and peanut brightness-greenness-humidity three-dimensionally over a time period T. BGW3DCC The optimal time window T for soybean and peanut identification was determined. Key .

4. The method for co-identification of soybeans and peanuts during the soybean-peanut season according to claim 1, characterized in that, In step S03, the target area is first preliminarily identified based on the three-dimensional coupling coordination index to screen candidate areas for leguminous crops. Then, a crop canopy coverage index is introduced within the candidate areas to further distinguish between soybeans and peanuts, obtaining high-confidence detection results. The detection rules are as follows: in, Based on the three-dimensional coupling coordination index With crop canopy coverage index The classification discriminant function outputs values ​​1, 2, and 3, corresponding to soybeans, peanuts, and other crops, respectively; the threshold... , The density distribution characteristics of field sample data were used to determine the data; confidence constraints were applied to randomly generated crop point samples to obtain positive samples of soybeans and peanuts with high confidence.

5. The method for co-identification of soybeans and peanuts during the soybean-peanut season according to claim 1, characterized in that, In step S04, an optimized feature set for identifying differences between soybeans and peanuts is constructed based on a combination of mechanistic understanding and data-driven approaches. First, heterogeneous features are extracted from optical and microwave remote sensing data, focusing on multidimensional attributes including crop growth characteristics, biochemical and physiological states, and canopy structure. Specifically, based on Sentinel-2 MSI, bands including Blue, Green, Red, red-edge, near-infrared, and short-wave infrared are selected to construct spectral index features reflecting vegetation growth, pigment content, soil exposure, and water status. Simultaneously, Sentinel-1 radar backscattering coefficients are introduced to extract microwave scattering features characterizing crop canopy structure and water content. Based on this, temporal statistical analysis is performed on the heterogeneous features throughout the crop's entire growth period, calculating the feature mean image to mitigate the impact of short-term temporal fluctuations and obtain a temporally stable original feature image set. Further, combined with field sample point data, the Jeffries–Matusita method is used... Distance is used to quantitatively evaluate the feature separability between target crop categories and non-target categories. Feature subsets with optimal discriminative ability for soybeans and peanuts are selected and optimized feature sets for soybeans and peanuts are constructed to provide highly discriminative and low-redundancy feature inputs for subsequent classification.

6. The method for co-identification of soybeans and peanuts during the soybean-peanut season according to claim 1, characterized in that, In step S05, a sample intelligent generation module combining confidence level and feature distribution dual constraints is constructed to achieve the automatic construction of large-scale, high-confidence positive and negative samples of soybeans and peanuts. Randomly generated crop point samples within the cultivated land mask area are used as a candidate sample set, and the candidate samples are screened and purified through positive and negative sample construction mechanisms respectively. Specifically, the positive sample construction mechanism achieves the automatic generation of highly reliable positive samples through the synergistic effect of confidence level constraints and spatial consistency constraints; the negative sample construction mechanism achieves the automatic generation of highly pure negative samples through the synergistic effect of feature distribution constraints and exponential interval constraints. Regarding positive sample construction, firstly, based on the high-confidence soybean and peanut detection results obtained in step S03, confidence constraints are applied to the randomly generated crop point samples to obtain initial candidate positive samples. Based on this, an S×S spatial neighborhood window is introduced to perform consistent convolution filtering on the spatial distribution of the candidate positive samples to suppress spatially isolated pixels and locally misjudged samples, obtaining soybean and peanut positive samples with good spatial continuity and high confidence. Regarding negative sample construction, based on step S04… A predetermined set of preferred features for soybeans and peanuts is used to construct a probability density distribution model for each feature dimension, based on real soybean and peanut samples. This model is then used to screen random crop point samples by applying feature distribution constraints. When a candidate sample simultaneously meets the distribution criteria of "non-soybean" and "non-peanut" across all feature dimensions, it is identified as a potential negative sample. Furthermore, by examining whether the three-dimensional coupling coordination index of the potential negative sample falls within the feature interval of non-soybean and non-peanut samples, a secondary constraint is applied to the negative sample to further improve its purity and reliability.

7. The method for co-identification of soybeans and peanuts during the soybean-peanut season according to claim 1, characterized in that, In step S06, a collaborative identification technology process for legume crops is constructed to obtain the post-season spatial distribution results of legume crops on a multi-year scale. To improve the generalization ability of the model under different year conditions, an interannual transfer enhancement strategy is introduced in the model training and application stages, and a system design is carried out from three levels: feature processing, model construction, and training strategy. At the feature processing level, phenological alignment and interannual normalization are performed on the multi-source temporal features of soybeans and peanuts from multiple source years to reduce feature distribution drift caused by factors including differences in phenological processes, changes in climate conditions, and inconsistencies in remote sensing observation conditions, thereby enhancing the statistical consistency of feature expressions from different years. At the model construction level, the year is introduced as an observation condition variable into the model training process. By imposing year-insensitive constraints, the model's dependence on year-difference information is limited, guiding the model to pay more attention to the stable growth characteristics and canopy structure differences of the crop itself, thereby learning crop discriminative representations with year invariance. At the model training level, the model is jointly trained based on high-confidence positive and negative sample data from multiple source years, so that the model is simultaneously constrained and optimized under multi-year conditions, resulting in a crop identification model with stable discrimination ability for different years. Finally, the crop identification model with cross-year transferability is directly applied to remote sensing image data of the target year, realizing the identification of the post-season spatial distribution of legumes on a multi-year scale without the need for samples from the target year to participate in training.

8. The method for co-identification of soybeans and peanuts during the soybean-peanut season according to claim 1, characterized in that, In step S07, to automatically construct reliable crop season identification training samples from stable planting areas at a multi-year scale, a spatiotemporal perception-based soybean and peanut distribution inference model is constructed. The inference model is implemented using a difference-in-difference model. The inference model is based on the multi-year scale legume distribution results obtained through the legume collaborative identification technology process constructed in step S06. It identifies high-confidence areas that maintain stable planting characteristics in time and space by combining temporal inference with spatial constraints. Taking the multi-year scale post-season spatial distribution results of legumes obtained through the legume collaborative identification technology process constructed in step S06 as input, it uses multi-year temporal information and spatial comparison relationships to weaken non-crop disturbances, including interannual climate fluctuations, sensor differences, and data noise, through a difference-in-difference model. Based on the parallel trend assumption, the inference model uses the temporal superposition of multi-year classification results to select pixels with similar spectral characteristics between T1 and T2 and no land cover change over many years to construct a control group for the difference-in-difference model. Subsequently, the differences in spectral and radar backscattering characteristics between the target pixel and the control group were calculated in the two time periods, and the true crop change signal was extracted through two difference operations; the calculation formula is as follows: Where x represents the feature value of the target pixel; μ and This represents the mean and covariance matrix of the control group's eigenvalues. This represents the Mahalanobis distance between the target pixel x and the control group H; , , and These represent the feature values ​​of the target pixel and the control group for years T1 and T2, respectively. and This represents the change before and after the change; the change before and after the change is subtracted a second time, and the absolute value is taken to obtain the change DD: The above formula characterizes the difference in the degree of change of the spectral and radar backscattering features between the target pixel and the control group from T1 to T2. Its purpose is to perform a second difference operation on the interannual variation to offset the influence of regional or systematic interannual variation. Based on the stable planting area over many years, a spatial neighborhood convolution strategy is further introduced to screen and constrain the spatial continuous area, and automatically generate a multi-year continuous training sample set of soybeans and peanuts to support the training and optimization of the crop season identification model.

9. The method for co-identification of soybeans and peanuts during the soybean-peanut season according to claim 1, characterized in that, In step S08, firstly, the Brightness-Greenness-Humidity Difference Index (BGWDI) is constructed to characterize the early spectral response of crops. Its calculation formula is as follows: in, , and They represent Brightness, Greenness, and Wetness values ​​at specific points in time. The moment when the first difference of the Greenness index becomes negative is defined as the moment when the first difference of the Greenness index becomes negative. The early identification window is formed by taking M days before and N days after that moment. Secondly, based on the analysis of stable soybean and peanut samples over many years, its Due to historical trajectories, soybeans and peanuts exhibit differences in canopy density in their early stages. A canopy density index (CDI) is constructed to characterize these differences in crop canopy structure and distinguish between soybeans and peanuts in their early stages. The calculation formula is as follows: in, T represents peak Number of observations within the time period; Then, in the BGWDI–CDI feature domain, each cell is represented as a two-dimensional feature vector. Based on years of stable training samples, the distribution characteristics of the crop in the feature space were analyzed. For different crop types c, the stable value ranges over many years were extracted in the BGWDI and CDI dimensions, and the corresponding stable feature spaces were constructed accordingly. Defined as: in, , and , These represent the lower and upper bounds of the multi-year stable values ​​of soybeans / peanuts in the BGWDI and CDI dimensions, respectively. The boundary parameters are determined by statistical analysis of the density distribution of field samples in the feature space. Finally, based on a stable feature space, a rule-based collaborative identification method for mid-season soybeans and peanuts is constructed; when the feature vector x corresponding to a pixel falls into the stable feature space of a certain crop... If x is not in the stable feature space of soybean / peanut, it is classified as the corresponding crop; if x is not in the stable feature space of soybean / peanut, it is classified as another crop; given that the stable feature spaces of soybean and peanut may overlap in local areas, a priority discrimination rule based on the center distance of the feature spaces is further introduced; the center position of soybean / peanut in the BGWDI–CDI feature space is: When a pixel feature vector x falls within the overlapping region of both the soybean and peanut feature spaces, the Euclidean distance between that pixel and the soybean and peanut feature spaces is calculated separately, and the minimum distance is taken as the discrimination type. By constructing feature space and establishing discrimination rules, we can achieve collaborative identification of mid-season soybeans and peanuts and reduce confusion between the two crops due to feature similarity during their critical growth periods.

10. A soybean-peanut season collaborative identification device, characterized in that, include: The system includes a time-series data processing module for optical radar, a three-dimensional coupling coordination index construction module, a multi-year continuous post-season mapping module for crops, a mid-season discrimination model construction module for soybeans and peanuts, and a mid-season collaborative identification result output module. The optical radar time-series data processing module is used to acquire and preprocess optical and radar remote sensing time-series data; on the remote sensing cloud platform, Sentinel-1 SAR and Sentinel-2 MSI image data that meet the research time period are selected, and image mosaicking, speckle filtering, cloud removal, cropping, interpolation, and temporal smoothing are performed respectively to generate optical and radar time-series image datasets with uniform temporal resolution; based on the processed Sentinel-2 MSI time-series images, a multi-dimensional spectral index time-series dataset including vegetation index, soil and moisture index, and pigment and browning index is further constructed for subsequent crop feature analysis and identification. The 3D Coupling Coordination Index Construction Module is used to construct a 3D coupling coordination index based on the simultaneous heterogeneity of brightness, greenness, and humidity according to the remote sensing time series characteristics of crops throughout their entire growth period, in order to characterize the synergistic response relationship of brightness, greenness, and humidity characteristics during the transition process of growth stages; the multi-year continuous post-season mapping module for crops identifies stable 3D coupling characteristics formed by crops in key growth stages by analyzing the time series changes of the 3D coupling coordination index, and automatically determines the optimal time window, thereby improving the stability and reliability of crop type identification; The multi-year continuous post-season mapping module for crops is used to construct a collaborative crop identification process with cross-year migration capabilities to obtain multi-year crop post-season spatial distribution results. This module employs an inter-annual migration enhancement strategy to perform phenological alignment and inter-annual normalization on the temporal features of multiple source years, and applies year-insensitive constraints during model training to improve the model's cross-year generalization ability. The module directly applies the crop identification model trained over multiple years to remote sensing image data of the target year, achieving continuous post-season mapping of crops without the need for samples from the target year. The soybean-peanut mid-season discrimination model construction module is used to distinguish between soybeans and peanuts in the early stages. Based on years of stable training samples, it analyzes the historical evolution of crop multidimensional characteristics, extracts stable temporal features of key growth stages, and constructs a feature space and discrimination rules for mid-season collaborative identification. This module constructs the Brightness-Greenness-Humidity Difference Index (BGWDI) and the Canopy Density Index (CDI) to characterize the stable differences of crops in spectral response and canopy structure, and automatically determines stable intervals in the BGWDI-CDI feature space. Combined with rule-based discrimination and overlap distance disambiguation strategies, it achieves reliable differentiation between soybeans and peanuts in the early stages and reduces the risk of confusion during key growth periods. Mid-season co-identification result output module: used to integrate and output the results of mid-season co-identification of soybeans and peanuts; By integrating high-confidence identification results obtained based on three-dimensional coupling coordination degree and crop canopy coverage index, an optimal feature set is constructed and a large-scale positive and negative sample is automatically generated to form a collaborative identification model with cross-year migration capability and output post-season distribution results at a multi-year scale. The crop multi-year continuous post-season mapping module further combines multi-year distribution results with crop spatiotemporal reasoning mechanism to dynamically obtain the latest year's training samples and extract multi-year stable feature space, realizing the integration and highly reliable output of early soybean and peanut identification results.

Images