Multi-crop yield prediction method and device based on remote sensing collaborative inversion
By integrating multi-source spectral data and high-resolution remote sensing images through the remote sensing collaborative inversion method, the problems of single remote sensing indicators, low prediction accuracy and fragmented links in the crop yield assessment system are solved, efficient and automated crop yield prediction is achieved, and the accuracy and timeliness of crop yield assessment are improved.
Patent Information
- Application Number
- CN202511120983.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-12
- Publication Date
- 2025-09-12
AI Technical Summary
The existing crop yield assessment system has problems such as single remote sensing indicators, low prediction accuracy, poor efficiency and timeliness, fragmented links, and lack of unified process integration, which makes it impossible to build a high-efficiency, high-precision, and automated full-process closed-loop prediction system.
A multi-crop yield prediction method based on remote sensing collaborative inversion is adopted. By acquiring multi-source remote sensing image data and reconstructing time series remote sensing data, deep learning and physical models are combined to achieve multi-level linkage of crop area identification, agricultural parameter inversion and yield prediction. The multi-level linkage full-process model is used to integrate multi-source spectral data and high-resolution remote sensing images to form a high-precision, high-efficiency, and automated integrated yield prediction solution.
It has achieved high-precision and high-timeliness in crop yield predictions, built a closed-loop system of automated data flow, model collaborative calculation and dynamic feedback of results, and improved the overall efficiency and accuracy of crop yield assessment.
Smart Images

Figure CN120633954A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of image processing technology, in particular to the field of remote sensing image technology, and specifically to a method and device for predicting multi-crop yields based on remote sensing collaborative inversion. Background Art
[0002] With the rapid development of remote sensing satellite observation capabilities, agricultural remote sensing monitoring using multi-source remote sensing data has become an important means of agricultural informatization. However, in practical applications, agricultural monitoring and yield assessment still face key challenges such as multi-source data heterogeneity, discontinuous time series, insufficient parameter inversion accuracy, and fragmented processes.
[0003] Specifically, the discontinuity problem of time-series remote sensing data: the current remote sensing monitoring system is highly dependent on medium- and high-resolution optical data (such as Sentinel-2, Landsat-8, GF-6, etc.), but these data are extremely susceptible to obstruction by cloud and rain weather during the growing season, resulting in the loss of remote sensing data during critical periods, affecting the timeliness and accuracy of downstream crop identification and parameter inversion; crop identification is strongly dependent on samples and time series: mainstream crop classification methods (such as classification methods based on random forests, support vector machines or convolutional neural networks) usually rely on complete time series NDVI / EVI data and a large number of ground-labeled samples, which limits the early application of the model and its cross-regional generalization ability. The inversion of agricultural parameters lacks physical mechanism support: most existing parameter inversion methods are based on empirical statistical relationships or a single vegetation index to construct inversion models, which lack unified physical mechanism support and are easily affected by changes in background conditions (such as soil moisture, leaf inclination, crop row spacing, etc.), and cannot stably adapt to multi-crop, multi-region, and multi-time scenarios; yield prediction models generally use statistical or single-point temporal methods: current yield assessments are mostly based on annual NDVI total values or single-month vegetation indices to construct regression models, which are difficult to depict the dynamic changes of multiple phases during the growth period and complex influencing factors such as extreme weather. At the same time, remote sensing inversion parameters are not effectively integrated, resulting in poor accuracy and insufficient timeliness.
[0004] Based on the above problems, the current international and domestic agricultural remote sensing yield prediction systems are mainly based on modularization or independent links. Representative solutions for approximate implementation methods include: USDA-NASS: This system relies on MODIS NDVI and statistical data, combined with ground surveys, to build statistical regression models to estimate yields. It has a low spatial resolution (250m) and lacks real-time updating and multi-parameter fusion capabilities. The EU JRC MARS Crop Yield Forecast System relies on meteorological data-driven crop models (such as WOFOST) combined with remote sensing indicators for regional crop monitoring and forecasting. However, the remote sensing and model systems are poorly coupled and lack high-resolution remote sensing parameter input. Research plans of domestic universities / research institutions: Most studies focus on a certain region (such as the North China Plain and the black soil in Northeast China) or a specific crop (such as wheat and corn). They use a single indicator (NDVI / EVI) in conjunction with random forest, support vector machine, SVR and other models to estimate yield. However, the overall process shows problems of fragmentation and discontinuity in the "identification-inversion-prediction" process, and there is a lack of unified interface standards and linkage mechanisms between modules.
[0005] However, in response to the core technical problems of current area measurement, namely the single remote sensing indicators, low prediction accuracy, poor efficiency and timeliness, fragmentation of key links, and lack of unified process integration in the current crop yield assessment process, the above methods often treat remote sensing data reconstruction, crop identification, agricultural parameter inversion and yield assessment as independent modules. For example, only optical remote sensing images are used for crop area identification, or single parameter inversion is relied upon on ground sampling. This approach lacks end-to-end data connectivity and system collaboration, making it difficult to build a high-efficiency, high-precision, and automated full-process closed-loop prediction system.
[0006] The above-mentioned major technical bottlenecks are as follows: 1. Fragmented links and poor coordination: Remote sensing data reconstruction, identification, inversion, and prediction are scattered across different platforms or algorithm systems. Data transmission relies on manual intervention, resulting in long processing cycles. The overall chain lacks closed-loop control, making it impossible to meet the needs of mid-season dynamic monitoring and rapid response. 2. Remote sensing indicators are isolated and models cannot share information: Parameters such as crop cover, leaf area index, canopy nitrogen content, canopy chlorophyll content, and aboveground biomass rely on different models for inversion, resulting in limited improvement in yield assessment accuracy and the inability to leverage the coupling characteristics between parameters for collaborative inversion. 3. Lack of integration and timeliness assurance mechanisms: Existing yield assessment solutions often use static parameter-driven models and fail to incorporate dynamic remote sensing time series processes. This results in significant prediction lags and makes it difficult to achieve "real-time tracking, dynamic adjustment, and precise control." 4. Low platform level and difficulty in technology transformation: Most of the current systems are decentralized algorithm demonstrations, and have not formed a universal and scalable system platform, and cannot be efficiently connected with domestic high-resolution remote sensing data or business monitoring systems.
[0007] In summary, the main shortcomings of current technologies include: Fragmented processes: Each link (data reconstruction, area identification, parameter inversion, and yield assessment) is independent and relies on manual coordination, resulting in delayed data transfer, accumulated errors, low overall process efficiency, and poor precision. The causal chain of technical defects can be summarized as frequent manual intervention leading to process disruptions, making closed-loop control impossible and resulting in weak business support capabilities. Limitations of time series data reconstruction: It often relies on single interpolation (such as linear interpolation) or filtering, failing to integrate historical priors with dynamic corrections to real-time data. When data is missing during periods of dense cloud and rain, the time series curve breaks, making it impossible to provide continuous and stable input for subsequent modules. The causal chain of these technical flaws can be summarized as the lack of a physical statistical fusion mechanism, resulting in low data reconstruction accuracy, which in turn affects the quality of the basic input for area identification and parameter inversion. Area identification lags: Relying on time-series data from the entire growing season (e.g., post-harvest inversion), this approach suffers from poor accuracy and timeliness in the early stages of the growing season, failing to meet the needs of real-time monitoring of agricultural production. The causal chain of its technical flaws can be summarized as the failure to combine phenological knowledge with deep learning using a small sample size, resulting in strong sample dependency and weak dynamic adaptability, leading to poor generalization of regional seasonal migration. Isolation in agricultural parameter inversion: Single-parameter inversion (e.g., LAI only) without coordination of multiple parameters (chlorophyll, nitrogen content, etc.), or purely empirical models ignoring physical mechanisms, result in inversion results that fail to reflect crop physiological synergies, resulting in low monitoring accuracy and poor adaptability throughout the entire growth period. The causal chain of technical defects can be summarized as the lack of physical model support and multi-parameter coordination, which leads to an inability to overcome interference from water and soil background and row sowing structure, resulting in weak inversion robustness across crops and growth periods. Extensive yield assessment: It relies on a single statistical model (such as regression) or simple machine learning, fails to integrate multi-source agricultural parameters and spatiotemporal characteristics, and cannot capture the interaction of complex environmental factors (such as the synergistic effects of accumulated temperature + precipitation + chlorophyll). The yield prediction accuracy is low and the timeliness is poor. The causal chain of its technical defects can be summarized as the simple model structure and insufficient feature utilization, which lead to the inability to adapt to climate change and regional heterogeneity, resulting in low reliability of yield assessment in major production areas. Summary of the Invention
[0008] The present disclosure provides a method, apparatus, device and storage medium for predicting multi-crop yields based on remote sensing collaborative inversion.
[0009] According to a first aspect of the present disclosure, a method for predicting multi-crop yields based on remote sensing collaborative inversion is provided. The method comprises: Obtain multi-source remote sensing image data of the area to be processed; Performing time series remote sensing data reconstruction processing on the multi-source remote sensing image data to generate a time series remote sensing data set; Based on a preset multi-crop area identification model, generating crop area identification data according to the time series remote sensing dataset; Based on a preset multi-agricultural condition multi-parameter inversion model, generating agricultural condition multi-parameter inversion data in the area to be processed according to the time series remote sensing data set and the crop area identification data; the agricultural condition multi-parameter inversion data includes leaf area index data, crop cover data, canopy chlorophyll data, canopy nitrogen content data, and aboveground biomass data; Based on the preset multi-crop yield prediction model, crop product prediction data is generated according to the agricultural multi-parameter inversion data.
[0010] According to the above aspects and any possible implementation, there is further provided an implementation, wherein the multi-source remote sensing image data includes historical multispectral data and current season multispectral data; The performing time series remote sensing data reconstruction processing on the multi-source remote sensing image data to generate a time series remote sensing data set includes: Based on the ground feature change detection algorithm and the harmonic wave function fitting algorithm, a historical reference curve is constructed according to the historical spectral data; Based on the extended Kalman filter prediction update mechanism, a time series remote sensing data set is generated according to the historical reference curve and the multispectral data of the current season.
[0011] According to the above aspects and any possible implementation, there is further provided an implementation, wherein the preset multi-crop area recognition model includes a deep learning driving module and a knowledge rule guiding module; The deep learning driving module adopts a twin network structure for single sample identification of each crop type; the knowledge rule guidance module constructs a crop identification decision tree for generating the identification area of each crop based on a preset standard time spectrum curve library for each crop and the time series remote sensing dataset.
[0012] According to the above aspects and any possible implementation, there is further provided an implementation, wherein the preset multi-agricultural condition multi-parameter inversion model includes a physical model module and a machine learning module; The physical model module includes the PROSAIL model and the CropGORT model, which are used to extract spectral features based on the time-series remote sensing dataset and the crop area identification data; the machine learning module is used to optimize nonlinear relationships through random forests and generate multi-parameter inversion data of agricultural conditions in the area to be processed based on the spectral features.
[0013] According to the aspects and any possible implementation methods described above, an implementation method is further provided, in which the preset multi-crop yield prediction model adopts a hybrid learning framework, including a back propagation network BP, a long short-term memory network and a convolutional neural network hybrid network model LSTM-CNN, which is used to systematically integrate the multi-parameter inversion data of the agricultural conditions to generate crop product prediction data.
[0014] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein the method further includes: The agricultural multi-parameter inversion data is subjected to spatial integration processing, dimension unification processing and variable redundancy removal processing.
[0015] According to a second aspect of the present disclosure, a multi-crop yield prediction device based on remote sensing collaborative inversion is provided. The device comprises: A data acquisition module is used to acquire multi-source remote sensing image data of the area to be processed; A data reconstruction module is used to perform time series remote sensing data reconstruction processing on the multi-source remote sensing image data to generate a time series remote sensing data set; an area identification module, configured to generate crop area identification data based on a preset multi-crop area identification model and the time series remote sensing dataset; An agricultural condition inversion module is configured to generate agricultural condition multi-parameter inversion data within the area to be processed based on a preset multi-agricultural condition multi-parameter inversion model and the time series remote sensing dataset and the crop area identification data; the agricultural condition multi-parameter inversion data includes leaf area index data, crop cover data, canopy chlorophyll data, canopy nitrogen content data, and aboveground biomass data; The yield prediction module is used to generate crop product prediction data based on a preset multi-crop yield prediction model and the agricultural condition multi-parameter inversion data.
[0016] Regarding the aspects described above and any possible implementation method, an implementation method is further provided, in which the data acquisition module, the data reconstruction module, the area identification module, the agricultural condition inversion module and the yield prediction module are a chain execution architecture, using the GeoTIFF format and an automatic data triggering mechanism, and the modules are automatically connected through a scheduling engine.
[0017] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the program.
[0018] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method described above is implemented.
[0019] The embodiment of the present application provides a multi-crop yield prediction method based on remote sensing collaborative inversion, which can obtain multi-source remote sensing image data of the area to be processed; then perform time series remote sensing data reconstruction processing on the multi-source remote sensing image data to generate a time series remote sensing data set; then generate crop area identification data based on the time series remote sensing data set based on a preset multi-crop area identification model; then generate agricultural condition multi-parameter inversion data in the area to be processed based on the time series remote sensing data set and the crop area identification data based on a preset multi-agricultural condition multi-parameter inversion model; the agricultural condition multi-parameter inversion data includes leaf area index data, crop coverage data, canopy chlorophyll data, canopy area identification data, and crop area identification data. Layer nitrogen content data and aboveground biomass data; then, based on the preset multi-crop yield prediction model, according to the agricultural multi-parameter inversion data, crop product prediction data is generated; based on this, a multi-level linkage full-process model of "time series remote sensing data reconstruction + crop area identification + agricultural parameter inversion + crop yield prediction" can be constructed. By integrating multi-source spectral data and high-resolution remote sensing images, unifying the inversion framework and data structure, and utilizing regional-level machine learning and deep learning models, while ensuring prediction accuracy, it realizes data flow automation, model collaborative calculation and dynamic feedback of results, forming a high-precision, high-efficiency, automated integrated yield prediction solution.
[0020] It should be understood that the contents described in the Summary of the Invention section are not intended to limit the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are provided for a better understanding of the present disclosure and do not constitute a limitation of the present disclosure. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which: Figure 1 A flowchart of a multi-crop yield prediction method based on remote sensing collaborative inversion according to an embodiment of the present disclosure is shown; Figure 2 A schematic diagram of the twin network structure of the deep learning driver module according to an embodiment of the present disclosure is shown; Figure 3 shows a structural diagram of an LSTM unit according to an embodiment of the present disclosure; Figure 4 A schematic diagram of an LSTM-CNN model according to an embodiment of the present disclosure is shown; Figure 5 A schematic diagram of corn yield accuracy inspection according to an embodiment of the present disclosure is shown; Figure 6 A schematic diagram of soybean yield accuracy inspection according to an embodiment of the present disclosure is shown; Figure 7 A schematic diagram of wheat yield accuracy inspection according to an embodiment of the present disclosure is shown; Figure 8 A schematic diagram of rice yield accuracy inspection according to an embodiment of the present disclosure is shown; Figure 9 shows a schematic diagram of an original image of a planting area according to an embodiment of the present disclosure; Figure 10 FIG4 shows a schematic diagram of a reconstructed image of a planting area according to an embodiment of the present disclosure; Figure 11 A wheat yield distribution map of region A according to an embodiment of the present disclosure is shown; Figure 12 shows a rice yield distribution map of region B according to an embodiment of the present disclosure; Figure 13 A corn yield distribution map of region C according to an embodiment of the present disclosure is shown; Figure 14 A soybean yield distribution map of region D according to an embodiment of the present disclosure is shown; Figure 15 A block diagram of a multi-crop yield prediction device based on remote sensing collaborative inversion according to an embodiment of the present disclosure is shown; Figure 16 shows a full-process architecture diagram according to an embodiment of the present disclosure; Figure 17 A block diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. DETAILED DESCRIPTION
[0022] To make the purpose, technical solutions, and advantages of the embodiments of the present disclosure more clear, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present disclosure.
[0023] In this document, the term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the related objects are in an "or" relationship.
[0024] In this disclosure, a multi-level linkage full-process model of "time series remote sensing data reconstruction + crop area identification + agricultural parameter inversion + crop yield prediction" can be constructed. By integrating multi-source spectral data and high-resolution remote sensing images, unifying the inversion framework and data structure, and utilizing regional-level machine learning and deep learning models, while ensuring prediction accuracy, it can realize data flow automation, model collaborative calculation and dynamic feedback of results, forming a high-precision, high-efficiency, automated integrated yield prediction solution.
[0025] Figure 1 A flowchart of a multi-crop yield prediction method 100 based on remote sensing collaborative inversion according to an embodiment of the present disclosure is shown.
[0026] In block 110 , multi-source remote sensing image data of the area to be processed is acquired.
[0027] In some embodiments, the multi-source remote sensing image data may include Sentinel-2 time-series satellite image data and Gaofen-6 data.
[0028] In some embodiments, using multi-source remote sensing image data as raw data can solve the problem of missing optical remote sensing data caused by poor observation conditions such as clouds and rain during the crop growing season, reconstruct continuous, stable, high-resolution time series remote sensing data, and provide basic support for crop identification, parameter inversion and yield assessment.
[0029] In some embodiments, the start time, end time, and region of multi-source remote sensing image data may be limited to more accurately generate a time series remote sensing dataset.
[0030] In block 120 , time series remote sensing data reconstruction processing is performed on the multi-source remote sensing image data to generate a time series remote sensing dataset.
[0031] In some embodiments, a time series remote sensing data reconstruction process based on historical and seasonal multispectral data can be constructed, with the start time, end time, and region as inputs, to output a time series remote sensing dataset.
[0032] In some embodiments, the multi-source remote sensing image data includes historical multispectral data and current season multispectral data; The above-mentioned reconstruction of multi-source remote sensing image data to generate a time series remote sensing dataset includes: Based on the ground feature change detection algorithm and the harmonic function fitting algorithm, a historical reference curve is constructed according to the historical spectral data; Based on the extended Kalman filter prediction update mechanism, a time series remote sensing dataset is generated according to the historical reference curve and the multispectral data of the current season.
[0033] In some embodiments, the process of constructing the historical reference curve specifically includes: The temporal stability of historical multispectral data was analyzed using the CCD algorithm; Identify representative ground feature stages based on the analysis results; The Normalized Difference Vegetation Index (NDVI) time series was smoothed and declouded using the harmonic function fitting (HANTS) algorithm to generate a benchmark vegetation time series change curve, also known as the historical reference curve.
[0034] In some embodiments, the process of Extended Kalman Filter (EKF) data assimilation specifically includes: First, the historical reference curve is used as the prior input, and the EKF prediction update mechanism is used to integrate near-real-time observation data, such as the multispectral data of the current season, for dynamic correction to simulate the time series change trend close to the real one, and obtain the multispectral reconstruction result at the current time point, that is, the time series remote sensing dataset at the current time point.
[0035] Among them, the EKF prediction update mechanism specifically includes: (1); (2); Among them, formula (1) represents the predicted state, formula (2) represents the variance estimation; represents the prior estimate of the current state, represents the prior error covariance, A represents the state transition matrix, B represents the control matrix, and Q represents the covariance matrix.
[0036] Secondly, the time series remote sensing dataset at the current time point is iteratively updated in real time. The prior estimate and its uncertainty are calculated through a linear combination of the prior model estimate and the weighted difference between the observation value and the model state prior estimate, and are updated according to the new real-time observation results.
[0037] The weights are defined by the "Kalman gain" (K), which includes: (3); (4); (5); Among them, formula (3) represents the calculation of Kalman gain, formula (4) represents the posterior calculation, and formula (5) represents the variance transfer; in the measurement update equation, the time k is the estimated value of x at time k, so that we can finally obtain a near real-time (NRT) value; and are all required for k+1 (future) estimates; the calculated Kalman gain ( ) is required for the current NRT estimate, which is also the most important part of this system of equations; k , represents the real-time observation value; H combines the state and the observation value H represents the real state ( ) is converted to a measured value ( ) is set to 1.
[0038] In some embodiments, the generation of a time series remote sensing dataset, i.e., near real-time NDVI and multispectral product generation, may further include: Based on the credibility estimation and temporal trend judgment of the NDVI vegetation index, the EKF method is used to re-estimate multi-band data to achieve near-real-time data output with high temporal resolution.
[0039] In block 130 , crop area identification data is generated based on a preset multi-crop area identification model and the time series remote sensing dataset.
[0040] In some embodiments, the preset multi-crop area recognition model can be specifically set according to the actual needs of the user.
[0041] In some embodiments, a preset multi-crop area identification model can be constructed based on a dual-track fusion identification mechanism of "knowledge rule guidance + deep learning driven", taking the time series remote sensing data file path as input and outputting the crop area identification Tiff file, so as to achieve high-precision, strong temporal and spatial generalization remote sensing area identification of major crops such as corn, wheat, rice, and soybeans in the absence of sufficient ground samples and early growing season data, especially improving the effectiveness and automation of early crop identification.
[0042] In some embodiments, the preset multi-crop area recognition model includes a deep learning driving module and a knowledge rule guiding module; The deep learning driven module adopts a twin network structure for single sample identification of various crop types; the knowledge rule guided module constructs a crop identification decision tree, which is used to generate the identification area of each crop based on the preset standard time spectrum curve library of each crop and the time series remote sensing dataset.
[0043] In some embodiments, the knowledge rule guidance module is used to realize automatic generation and update of knowledge-driven samples.
[0044] Specifically, the upper envelope constrained Whittaker smoothing (UE-WS) method is used to process time series images, that is, time series remote sensing datasets, to extract key phenological periods and characteristic indices; multi-indicator combination rules such as NDVI, Enhanced Vegetation Index (EVI), and Red Edge Normalized Difference Vegetation Index (RENDVI) are constructed, and combined with expert knowledge to generate a crop identification decision tree, realizing rapid and automated mapping of crop identification areas.
[0045] The calculations include NDVI, EVI, Land Surface Water Index (LSWI), RENDVI, and RENDVI2 indices, as shown below: ; ; ; ; .
[0046] In some embodiments, three selected bands, namely, near infrared band NIR, short wave infrared 1 SWIR1 and red band, can be selected to display RGB images in a 98% linear stretching manner to achieve a better display effect as in ArcGIS software, making it easier to distinguish between corn, rice, soybeans and wheat.
[0047] In some embodiments, statistical analysis can also be performed on the spectrum / vegetation index time series to form a standard time spectrum curve library for each crop and support cross-regional migration applications.
[0048] In some embodiments, the deep learning driving module is used to achieve accurate recognition of deep learning models.
[0049] Specifically, the Siamese Network structure is used to achieve single-sample recognition of crop types, and the SiaLSTM model and SiaGRU model are used to enhance the generalization ability across regions and crop types. Figure 2 As shown in the figure, multi-temporal Sentinel-2 optical band images are input, and VV / VH polarization, RE1~RE4, NDVI and other features are integrated to form a crop differentiation feature space.
[0050] Furthermore, a three-dimensional feature tensor input deep model can be constructed. The three dimensions are temporal generalization ability research, spatial generalization ability research, and crop generalization ability research. Feature similarity is used to measure the similarity of crop categories between different pixels to improve the model's adaptability and recognition accuracy.
[0051] In box 140, based on the preset multi-agricultural condition multi-parameter inversion model, agricultural condition multi-parameter inversion data in the area to be processed is generated according to the time series remote sensing data set and the crop area identification data; the agricultural condition multi-parameter inversion data includes leaf area index data, crop cover data, canopy chlorophyll data, canopy nitrogen content data and aboveground biomass data.
[0052] In some embodiments, the preset multi-agricultural condition multi-parameter inversion model can be specifically set according to the actual needs of the user.
[0053] In some embodiments, the preset multi-agricultural condition and multi-parameter inversion model can be a universal inversion framework that integrates physical models and machine learning to meet the inversion needs of key crop growth period parameters. It takes the time series remote sensing data file path, crop area identification Tiff file path, and area as input, and outputs leaf area index Tiff file, crop coverage Tiff file, canopy chlorophyll Tiff file, canopy nitrogen content Tiff file, and aboveground biomass Tiff file, so as to realize the inversion of key agricultural condition parameters such as leaf area index, crop coverage, canopy chlorophyll, canopy nitrogen content, and aboveground biomass, thereby overcoming the interference of water and soil background, supporting dynamic monitoring of crops throughout the growth period before and after ridge closure, and adapting to the differences in row sowing structure and physiological characteristics of wheat, rice, corn, and soybeans.
[0054] In some embodiments, the preset multi-agricultural condition multi-parameter inversion model includes a physical model module and a machine learning module; The physical model module includes the PROSAIL model and the CropGORT model, which are used to extract spectral features based on time-series remote sensing datasets and crop area identification data; the machine learning module is used to optimize nonlinear relationships through random forests and generate multi-parameter inversion data of agricultural conditions in the area to be processed based on spectral features.
[0055] In some embodiments, the pre-set multi-agricultural parameter inversion model can utilize a "physical model + machine learning" fusion inversion strategy, integrating the PROSAIL model and the CropGORT model to provide mechanistic support. Machine learning is then used to address nonlinear errors, enabling the coordinated inversion of key agricultural parameters. For ease of illustration, the following will specifically illustrate the inversion of leaf area index data, canopy chlorophyll data, and canopy nitrogen content data.
[0056] Specific methods include: Based on the PROSAIL model, sensitivity analysis is performed to identify the main influencing factors under different crop types, lock the sensitive spectral range, and optimize the weight coefficient in real time through the LSTM algorithm. Parameters such as leaf inclination angle (ALA) and soil cover ratio (FCsoil) are optimized to comprehensively characterize the multi-factor interaction effect and construct the leaf area index inversion. The LSTM unit structure diagram is shown in the figure below. Figure 3 As shown, where h in the legend represents the hidden state, represents the Sigmoid layer, and tanh represents the Tanh layer.
[0057] The CropGORT model is used to simulate the canopy spectrum. Combined with the row structure parameters of row crops, such as ridge width and row spacing, the canopy reflectance spectra under different planting scenarios are simulated. The nonlinear relationship is optimized through random forest to achieve "physical simulation + data-driven" error correction.
[0058] For the early and late stages of crop growth, an improved red-edge chlorophyll index ( ) and enhanced canopy chlorophyll index (ECI) to alleviate the interference of factors such as water and soil background and dry matter.
[0059] Using the reflectivity ratio of 477nm to 677nm ( / ) as a correction factor to adjust the red edge band (780nm and 720nm, i.e. / ) reflectivity difference, the formula is: Based on the Sentinel-2 red edge band (704nm, 740nm, 783nm, i.e. ), combined with the band combination with the strongest correlation with the semi-empirical model (CCC), the formula is: ;based on The semi-empirical model with visualized parameters and easy to use is constructed with ECI. The formula is: , ,in, 、 、 、 are model parameters respectively; the canopy dry matter content (CDM) is obtained by multiplying the dry matter content by the leaf area index. Therefore, the canopy nitrogen content (CNC) can be estimated from CCC and CDM using the formula: ,in, is the photosynthetic nitrogen distribution coefficient, is the non-photosynthetic nitrogen distribution coefficient.
[0060] In block 150 , crop product prediction data is generated based on a preset multi-crop yield prediction model and agricultural condition multi-parameter inversion data.
[0061] In some embodiments, the preset multi-crop yield prediction model can be specifically configured according to the actual needs of the user.
[0062] In some embodiments, the preset multi-crop yield prediction model can be aimed at the core problems of low yield prediction accuracy and poor timeliness in the main crop producing areas. It adopts a hybrid learning framework and combines the back propagation network (BP), long short-term memory network and convolutional neural network hybrid network model (LSTM-CNN) to construct it. The leaf area index Tiff file path, crop cover Tiff file path, canopy chlorophyll Tiff file path, canopy nitrogen content Tiff file path and aboveground biomass Tiff file path are used as input, and the yield evaluation Tiff file is output to improve the modeling ability of complex nonlinear relationships. It systematically integrates factor data such as vegetation cover, leaf area index, canopy chlorophyll, canopy nitrogen content, aboveground biomass, establishes spatial adaptation, time transfer learning and spatiotemporal collaborative modeling capabilities, and ultimately achieves high-resolution and generalizable yield product generation.
[0063] In some embodiments, the above-mentioned preset multi-crop yield prediction model adopts a hybrid learning framework, including a back propagation network BP, a long short-term memory network and a convolutional neural network hybrid network model LSTM-CNN, which is used to systematically integrate the agricultural multi-parameter inversion data to generate crop product prediction data.
[0064] In some embodiments, the construction of the preset multi-crop yield prediction model can be divided into a machine learning modeling mechanism and a deep neural network modeling mechanism.
[0065] In some embodiments, for the machine learning modeling mechanism, the features and model selection of the machine learning modeling mechanism can be to input the screened features into the random forest (RF) and gradient boosting tree (GBDT) models respectively, and build regional optimal models according to different counties (districts); the time transfer learning mechanism of the machine learning modeling mechanism can be based on the LAI, FVC, CCC, CNC, AGB and other feature prediction results of agricultural condition inversion parameters, and the model time series can be migrated forward to realize early prediction of yield.
[0066] In some embodiments, for the deep neural network modeling mechanism, the back propagation neural network (BP) can automatically fit the nonlinear relationship between complex yield and environmental factors, using a sliding window, such as time step tN to t, inputting a historical multi-year monthly indicator sequence, and predicting the yield in year t+1; in the first hidden layer: the input is mapped to the hidden size dimensional space, ReLU activation; the second layer: keeping the dimension unchanged, continuing feature extraction; the output layer: single neuron linear mapping, the prediction target is crop yield (kg / hm²).
[0067] The structure of the LSTM-CNN hybrid network is as follows Figure 4 As shown, it integrates the spatial feature modeling capabilities of CNN and the time series modeling capabilities of LSTM to improve the expression of complex dynamics and extreme weather impacts. The input is the time series of multiple indicators during the crop growth period of each county over the years. The input dimension structure supports multi-time step, multi-feature, and multi-county parallel processing. 1D convolution is used to extract local spatial features. The pooling layer compresses redundant information and improves noise resistance. The LSTM module (time-dependent modeling) models interannual changes and the lag effects of extreme events. The gating mechanism (input gate / forget gate) dynamically screens signals during key periods. The fusion layer splices and integrates the output features of the CNN and LSTM modules, and outputs the predicted value through regression through the FC layer.
[0068] In some embodiments, in order to improve the accuracy of crop product prediction data, data preprocessing and feature engineering can be performed on the agricultural multi-parameter inversion data, such as spatial integration processing, unified dimension processing, and variable redundancy removal processing.
[0069] For example, spatial integration: superimpose with geographic boundaries to calculate the monthly cumulative value / average of county (district) level units, such as total NDVI value, average temperature, precipitation, etc.; unified dimension processing: use Z-score standardization for all environmental factors to improve the ability to compare features and reduce the impact of physical quantity differences; variable redundancy: through correlation coefficient screening (threshold 0.8), remove highly redundant variables and improve the generalization ability of the model.
[0070] According to the embodiments of the present disclosure, the following technical effects are achieved: By acquiring multi-source remote sensing image data of the area to be processed; then reconstructing the multi-source remote sensing image data to generate a time series remote sensing data set; then based on a preset multi-crop area recognition model, generating crop area recognition data according to the time series remote sensing data set; then based on a preset multi-agricultural condition multi-parameter inversion model, generating agricultural condition multi-parameter inversion data in the area to be processed according to the time series remote sensing data set and the crop area recognition data; the agricultural condition multi-parameter inversion data includes leaf area index data, crop coverage data, canopy chlorophyll data, canopy nitrogen content data and aboveground biomass data; then Based on the preset multi-crop yield prediction model, crop product prediction data is generated according to the agricultural multi-parameter inversion data; based on this, a multi-level linkage full-process model of "time series remote sensing data reconstruction + crop area identification + agricultural parameter inversion + crop yield prediction" can be constructed. By integrating multi-source spectral data and high-resolution remote sensing images, unifying the inversion framework and data structure, and utilizing regional-level machine learning and deep learning models, while ensuring prediction accuracy, it realizes data flow automation, model collaborative calculation and dynamic feedback of results, forming a high-precision, high-efficiency, automated integrated yield prediction solution.
[0071] In some embodiments, to verify the present disclosure, the following will be elaborated in the manner of technical verification.
[0072] In some embodiments, the coefficient of determination ( ), root mean square error (RMSE), normalized root mean square error (NRMSE), and mean relative error (MRE) can be selected as the accuracy verification indicators for the single - yield estimation results. The root mean square error (RMSE) on the training set after training is calculated to evaluate the performance and fitting effect of the current model.
[0073] In some embodiments, the root mean square error measures the average error magnitude between the predicted value and the true value, and the formula is: ; where is the true value, is the predicted value, and is the number of samples.
[0074] In some embodiments, relative accuracy can be used to represent the error between the true value and the predicted value. The calculation formula for relative accuracy is: ; where is the true value, is the predicted value, and the closer the result is to 1, the higher the prediction accuracy.
[0075] In some embodiments, the ground - sampled true value is converted into the yield at the pixel (10×10) scale, and the relative accuracy between the true value and the predicted value is calculated. The single - yield accuracy test for each crop is as Figure 5-Figure 8 shown. Among them, the coefficient of determination of the corn single - yield estimation model [[ID=,32]]is 0.65, RMSE is 606.30 kg / ha, NRMSE is 10.92%, and MRE is 9.56%; the coefficient of determination of the soybean single - yield estimation model is 0.38, RMSE is 211.48 kg / ha, NRMSE is 10.59%, and MRE is 8.74%; the RMSE of the winter wheat single - yield is 562.54 kg / ha, NRMSE is 9.94%, and the accuracy can reach more than 90%; the RMSE of the rice single - yield is 656.55 kg / ha, NRMSE is 9.66%, and the accuracy can reach more than 90%. ,
[0076] In some embodiments, the present disclosure forms the instant data production of crops in the large - area growing season. Taking the main crop - planting areas as an example, the comparison schematic diagrams before and after reconstruction with the images from March to May (10 - day period / standard false - color synthesis) are as Figure 9 and Figure 10 shown. Among them, Figure 9 is the original image of a certain area from March to May in 2024, and its geographical layer review number is GS(2024)1491 - A - Survey Qualification No. 34100640; Figure 10It is the reconstructed image of a certain area from March to May 2024, and its geographical layer review number is GS(2024) No. 1491 - Class A Survey Qualification No. 34100640.
[0077] In some embodiments, the present disclosure can accurately capture the single - yield distribution of wheat, rice, corn, and soybeans, and apply the trained model to the main crop - producing areas, enabling crop yield estimation with strong timeliness and high spatial resolution. The crop yield estimation results in some areas are as Figure 11-14 shown. Among them, Figure 11-14 only the crop yield estimation distribution status is shown, which is a crop yield estimation sample layer obtained based on geographical regions.
[0078] In summary, due to the fact that remote - sensing data is often blocked by clouds and rain during the crop growth season, the present disclosure solves the problem of incomplete traditional remote - sensing time series through the historical curve + EKF assimilation method, significantly improving timeliness and stability; because crop recognition highly depends on a large number of labeled samples, the present disclosure introduces knowledge - driven + siamese network, enabling high - precision recognition even in small - sample scenarios and being transferable to different crop belts; because the inversion of agricultural situation parameters is easily interfered by soil and water backgrounds if relying on a single spectral feature, the present disclosure enhances the inversion stability in the early, middle, and late stages through the two - stage dynamic models of mCIred - edge and ECI; because crop yields have complex spatio - temporal non - linear characteristics and are difficult to fit with current statistical regression models, the present disclosure introduces a sliding - window LSTM - CNN fusion network, effectively capturing the dual characteristics of spatial distribution and time dynamics, and enhancing the generalization prediction ability of the model; based on this, after system integration, it has the ability of automatic task scheduling and module triggering, forming a highly coupled end - to - end business closed - loop, and solving the efficiency and accuracy bottlenecks caused by the fragmentation of traditional chains.
[0079] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present disclosure is not limited by the described action sequence, because according to the present disclosure, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present disclosure.
[0080] The above is the introduction of method embodiments. The following further illustrates the solution of the present disclosure through device embodiments. <000031 The data reconstruction module 1520 is used to perform time series remote sensing data reconstruction processing on multi-source remote sensing image data to generate a time series remote sensing data set; An area identification module 1530 is configured to generate crop area identification data based on a preset multi-crop area identification model and a time series remote sensing dataset; The agricultural condition inversion module 1540 is configured to generate agricultural condition multi-parameter inversion data for the processing area based on a preset multi-agricultural condition multi-parameter inversion model, according to the time series remote sensing dataset and the crop area identification data; the agricultural condition multi-parameter inversion data includes leaf area index data, crop cover data, canopy chlorophyll data, canopy nitrogen content data, and aboveground biomass data; The yield prediction module 1550 is used to generate crop product prediction data based on a preset multi-crop yield prediction model and multi-parameter inversion data of agricultural conditions.
[0082] In some embodiments, the above-mentioned data acquisition module, data reconstruction module, area identification module, agricultural condition inversion module and yield prediction module are chain execution architectures, using GeoTIFF format and automatic data triggering mechanism, and each module is automatically connected through a scheduling engine.
[0083] The whole process architecture is as follows Figure 16As shown. Specifically, the time series remote sensing data reconstruction module: in order to solve the problem of missing optical data caused by cloud and rain obstruction during the growing season, the dual-engine mechanism of "historical benchmark curve + dynamic assimilation of the current season" is adopted, and parameters are set to constrain the time series curve change space. Sentinel-2 time series satellite image data and Gaofen-6 data are used as data sources. Combined with the high-temporal and spatial resolution remote sensing data reconstruction technology of filtering and curve fitting, real-time data production is formed during the growing season of large crop areas. Crop area identification module: in view of the fact that the current crop area identification remote sensing method mainly relies on time series data collected throughout the growing season, the lag problem of traditional methods relying on the complete growing season time series is overcome, and a "knowledge rule guidance + deep learning driven" dual-track model is established; based on the reconstruction results of time series remote sensing data as input, the deep learning model based on a small number of samples is used to accurately identify crops, and the model is verified to have strong spatiotemporal generalization capabilities; agricultural parameter inversion module: breaking through the limitations of isolated parameter inversion, using the reconstruction results of time series remote sensing data and crop identification results as input, a unified PROSAIL-CropGORT framework is constructed to achieve Multi-parameter collaborative optimization, this model effectively couples multi-dimensional physical processes with machine learning methods, breaks through the limitations of traditional inversion technology, and provides a scientific theoretical basis and technical path for remote sensing monitoring of crop growth status; Crop yield prediction module: Aiming at the core problems of low accuracy and poor timeliness of crop yield estimation, with multi-source data fusion and intelligent algorithm innovation as the main line, it systematically integrates the results of agricultural parameter inversion, vegetation coverage, leaf area index, canopy chlorophyll, canopy nitrogen content, aboveground biomass and other factor data, and innovatively develops regional machine learning models and deep learning models for major crop production areas, effectively improving the accuracy and timeliness of yield prediction.
[0084] As can be seen, the various modules in the system not only interact with each other through standardized data interfaces, but also achieve process integration through an "automatic data triggering mechanism." That is, after each module completes calculations and outputs, the system automatically detects its data status and triggers downstream modules to perform corresponding processing tasks. This mechanism is based on rule triggering and scheduling engine control and has the following characteristics: Unified data standards: All module outputs conform to unified data specifications, such as GeoTIFF raster, supporting seamless connection between modules; Automatic task scheduling: After the module is executed, the downstream model is automatically called without manual intervention, ensuring that the entire data processing process is online; Model chain execution: Forming a logical chain from data reconstruction to yield assessment, each link provides key input for the other, avoiding information fragmentation; Abnormal backtracking capability: The system supports process interruption detection and breakpoint recovery to ensure process robustness and reliability.
[0085] Through the close connection mechanism of multi-level linkage and automatic triggering process control, the system realizes the closed-loop integrated management and control logic of "data-driven-identification-inversion-evaluation", providing efficient, intelligent and stable business support capabilities for the crop yield assessment in the main production areas.
[0086] In summary, the core innovation of this disclosure lies not only in the integrity and coordination of the remote sensing yield prediction process, but also in the systematic integration and multiple breakthroughs in data reconstruction mechanism, identification model structure, parameter inversion method, and yield modeling architecture. Specifically, it includes the following four aspects: 1. Architecture design of full-process closed-loop remote sensing monitoring system An innovative four-level modular chain execution architecture encompassing "data reconstruction - area identification - parameter inversion - yield assessment" is constructed. Through the standardized GeoTIFF format and an "automatic data triggering mechanism," an end-to-end closed loop from data input to yield output is achieved. All modules are automatically connected through a scheduling engine, enabling backtracking from abnormal interruptions and task resumption from breakpoints. This system adapts to the operational deployment requirements of large areas and multiple crops, ensuring system robustness, real-time performance, and collaboration. 2. Knowledge-driven + Deep Siamese Network Area Intelligent Recognition Strategy A crop recognition framework was designed that integrates phenological knowledge rules with a twin neural network structure. It automatically generates crop labels for a small number of samples and uses temporal feature similarity to identify regional migrations in a single sample. This overcomes the problems of traditional methods with strong sample dependence and high reliance on temporal integrity, improving the model's generalization capabilities across crops and regions. 3. Parameter coupling inversion mechanism for multi-physics model fusion The innovative integration of PROSAIL and CropGORT models models "spectral-physiological parameters" and "structure-reflectance mechanism," respectively. A collaborative inversion framework for multiple parameters, including CCC, CNC, LAI, FVC, and AGB, was constructed to overcome the drawbacks of isolated single-indicator modeling. All inversion models combine physical simulation with machine learning, such as random forests and LSTM, for error correction, ensuring interpretability and high generalization. 4. LSTM-CNN hybrid network driven spatiotemporal migration yield prediction model A deep learning prediction structure is constructed based on BP neural network, LSTM and 1D-CNN; the dynamic nonlinear relationship between "climate-agricultural conditions-yield" is modeled using sliding window input of historical environmental data; the CNN module extracts regional local spatial features, and LSTM captures inter-annual temporal dependencies; it supports spatial scale change adaptation and time migration learning from pixel level to county level, and its accuracy is significantly better than that of statistical regression models.
[0087] It should also be noted that as a technical alternative to time series data reconstruction, generative adversarial networks, such as GANs, can be used. For example, the generator is trained to learn the distribution of cloud-free time series data, the discriminator distinguishes between real and generated data, and the missing data is reconstructed through adversarial training. However, a large number of cloud-free samples are required for training, and the effect is limited when there is insufficient data for small-area scenes. In addition, the physical mechanism is difficult to integrate, and it is not as stable as the EKF combined with historical priors. In addition, if a spatiotemporal interpolation method is used, such as Kriging interpolation, the missing values are filled by interpolating spatially adjacent pixels and time trends. However, it relies on spatial correlation, and the accuracy is poor in areas with dense clouds and rain or areas with strong heterogeneity of land features, such as the farmland-urban boundary, and cannot replace the physical-statistical fusion mechanism disclosed in this disclosure.
[0088] As for alternative technologies for area recognition, current pure machine learning methods, such as random forests, rely on manual feature engineering and large numbers of samples. This can lead to insufficient data at the beginning of the growing season and poor generalization when migrating across regions, making it incapable of addressing the needs of small-sample, highly dynamic scenarios. Current pure semantic segmentation models, such as U-Net, require large numbers of labeled samples and are less adaptable to phenological variations and cross-regional changes in landforms, making them a poor substitute for the low-sample generalization capabilities of the "knowledge + deep learning" dual approach.
[0089] Regarding alternative technologies for inverting agricultural parameters, using current purely empirical statistical models, such as multivariate linear regression, ignores crop physiology and structural characteristics, such as the influence of row width. Inversion accuracy is limited by sample distribution and cannot be adapted to the entire growth period and multiple crop types. Using single physical models, such as PROSAIL alone, fails to consider the soil and water background and structural interference of row crops, such as canopy reflectance errors when LAI < 5. This results in poor inversion robustness and makes it difficult to replace the collaborative optimization of physical and machine learning integration.
[0090] The above methods have obvious deficiencies in accuracy, sample dependence or explanatory power, so the present disclosure provides an integrated strategy that is more adaptable, accurate and theoretically based.
[0091] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.
[0092] In the technical solutions disclosed herein, the acquisition, storage, and application of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0093] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.
[0094] Figure 17A block diagram of an exemplary electronic device 1700 capable of implementing embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are provided as examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.
[0095] Electronic device 1700 includes a computing unit 1701, which can perform various appropriate actions and processes according to a computer program stored in ROM 1702 or a computer program loaded from storage unit 1708 into RAM 1703. RAM 1703 may also store various programs and data required for the operation of electronic device 1700. Computing unit 1701, ROM 1702, and RAM 1703 are connected to each other via a bus 1704. An I / O interface 1705 is also connected to bus 1704.
[0096] Multiple components in electronic device 1700 are connected to I / O interface 1705, including: an input unit 1706, such as a keyboard, mouse, etc.; an output unit 1707, such as various types of displays, speakers, etc.; a storage unit 1708, such as a magnetic disk, optical disk, etc.; and a communication unit 1709, such as a network card, modem, wireless communication transceiver, etc. The communication unit 1709 allows electronic device 1700 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.
[0097] Computing unit 1701 can be various general-purpose and / or specialized processing components with processing and computing capabilities. Some examples of computing unit 1701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. Computing unit 1701 performs the various methods and processes described above, such as method 100. For example, in some embodiments, method 100 can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 1708.
[0098] In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 1700 via the ROM 1702 and / or the communication unit 1709. When the computer program is loaded into the RAM 1703 and executed by the computing unit 1701, one or more steps of the method 100 described above may be performed. Alternatively, in other embodiments, the computing unit 1701 may be configured to perform the method 100 in any other appropriate manner (e.g., via firmware).
[0099] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
[0100] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.
[0101] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fibers, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0102] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).
[0103] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.
[0104] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.
[0105] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.
[0106] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.
Claims
1. A multi-crop yield prediction method based on remote sensing collaborative inversion, characterized in that: include: Obtain multi-source remote sensing image data of the area to be processed; Performing time series remote sensing data reconstruction processing on the multi-source remote sensing image data to generate a time series remote sensing data set; Based on a preset multi-crop area identification model, generating crop area identification data according to the time series remote sensing dataset; Based on a preset multi-agricultural condition multi-parameter inversion model, generating agricultural condition multi-parameter inversion data in the area to be processed according to the time series remote sensing data set and the crop area identification data; the agricultural condition multi-parameter inversion data includes leaf area index data, crop cover data, canopy chlorophyll data, canopy nitrogen content data, and aboveground biomass data; Based on the preset multi-crop yield prediction model, crop product prediction data is generated according to the agricultural multi-parameter inversion data.
2. The method according to claim 1, characterized in that The multi-source remote sensing image data includes historical multispectral data and current season multispectral data; The performing time series remote sensing data reconstruction processing on the multi-source remote sensing image data to generate a time series remote sensing data set includes: Based on the ground feature change detection algorithm and the harmonic wave function fitting algorithm, a historical reference curve is constructed according to the historical spectral data; Based on the extended Kalman filter prediction update mechanism, a time series remote sensing data set is generated according to the historical reference curve and the multispectral data of the current season.
3. The method according to claim 2, characterized in that The preset multi-crop area recognition model includes a deep learning driving module and a knowledge rule guiding module; The deep learning driving module adopts a twin network structure for single sample recognition of each crop type; The knowledge rule guidance module constructs a crop identification decision tree for generating the identification area of each crop according to the time series remote sensing data set based on a preset standard time spectrum curve library of each crop.
4. The method according to claim 3, characterized in that The preset multi-agricultural condition multi-parameter inversion model includes a physical model module and a machine learning module; The physical model module includes a PROSAIL model and a CropGORT model, which are used to extract spectral features based on the time series remote sensing dataset and the crop area identification data; The machine learning module is used to optimize nonlinear relationships through random forests and generate multi-parameter inversion data of agricultural conditions in the area to be processed based on the spectral characteristics.
5. The method according to claim 4, characterized in that The preset multi-crop yield prediction model adopts a hybrid learning framework, including a back propagation network BP, a long short-term memory network and a convolutional neural network hybrid network model LSTM-CNN, which is used to systematically integrate the agricultural multi-parameter inversion data to generate crop product prediction data.
6. The method according to any one of claims 1 to 5, characterized in that The method further comprises: The agricultural multi-parameter inversion data is subjected to spatial integration processing, dimension unification processing and variable redundancy removal processing.
7. A multi-crop yield prediction device based on remote sensing collaborative inversion, characterized in that: include: A data acquisition module is used to acquire multi-source remote sensing image data of the area to be processed; A data reconstruction module is used to perform time series remote sensing data reconstruction processing on the multi-source remote sensing image data to generate a time series remote sensing data set; an area identification module, configured to generate crop area identification data based on a preset multi-crop area identification model and the time series remote sensing dataset; An agricultural condition inversion module is configured to generate agricultural condition multi-parameter inversion data within the area to be processed based on a preset multi-agricultural condition multi-parameter inversion model and the time series remote sensing dataset and the crop area identification data; the agricultural condition multi-parameter inversion data includes leaf area index data, crop cover data, canopy chlorophyll data, canopy nitrogen content data, and aboveground biomass data; The yield prediction module is used to generate crop product prediction data based on a preset multi-crop yield prediction model and the agricultural condition multi-parameter inversion data.
8. The device according to claim 7, characterized in that The data acquisition module, the data reconstruction module, the area identification module, the agricultural situation inversion module and the yield prediction module are chain execution architectures, using the GeoTIFF format and automatic data triggering mechanism, and the modules are automatically connected through a scheduling engine.
9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 6.
10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-6.
Citation Information
Patent Citations
Winter wheat identification method based on hyperspectral data
CN115690582A
Crop yield inversion method and system based on fusion network model
CN118735140A
Near-real-time reconstruction method, device, equipment and medium for multispectral remote sensing time sequence images in whole growth season of crops
CN119495023A
Crop yield prediction method based on time sequence remote sensing image
CN120278341A
Crop monitoring system and method based on multispectral remote sensing and deep learning
CN120372217A
Cited By
Crop yield prediction method and system integrating multi-source remote sensing data
CN120976785A
Multi-temporal remote sensing image fusion method and device, equipment and storage medium
CN121214258A
Crop yield prediction method based on data enhancement and transfer learning technology
CN121303434A
Crop yield parameter inversion method and system based on mechanism constraint and data driving cooperation
CN122451428A
