Method and system for yield prediction based on crop growth model

Abstract

This application discloses a method and system for yield prediction based on a crop growth model. The method includes: obtaining the full growth curve of the target crop using remote sensing data of the target crop's production area from the planting time to the current time using a pre-trained vegetation index growth prediction model; generating heterogeneous development curves and heterogeneous guidance curves of the target crop using a preset spatial heterogeneity analysis mechanism, performing differential analysis with preset dual-benchmark curves, and extracting the target deviation interval; extracting the NDVI multivariate feature vector of the target crop's full growth curve based on the target deviation interval, inputting it into the pre-trained crop yield prediction model, and outputting the predicted yield value of the target crop. This comprehensively improves the accuracy and relevance of crop yield prediction.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing technology, and in particular to a method and system for yield prediction based on crop growth models. Background Technology

[0002] Accurate crop yield forecasting is crucial for agricultural management departments and policymakers in formulating harvesting and storage plans, as well as preparing for food imports and exports. However, traditional methods, based on ground or experimental data, require repetitive labor for different crops, resulting in high labor costs and time consumption. Furthermore, they are difficult to cover large areas of farmland, especially in regions with variable climates and complex environments.

[0003] With the development of remote sensing technology, it has become possible to predict crop yield using vegetation indices (such as NDVI). However, how to effectively utilize remote sensing data and combine it with crop growth models for accurate yield prediction remains an unsolved problem. Furthermore, existing technologies focus on the variation curves throughout the entire crop growth cycle. However, crop growth is significantly affected by spatial heterogeneity, such as differences in soil moisture and fertility, which are key factors influencing yield. Therefore, it is difficult to accurately capture the impact of key factors such as environmental stress gradients on yield prediction, limiting the accuracy and real-time nature of yield forecasts and easily leading to significant prediction errors. Summary of the Invention

[0004] This application provides a yield prediction method and system based on crop growth models. By combining crop growth models, spatial heterogeneity analysis, and dual benchmark curve comparison, it achieves dynamic and accurate yield prediction within the crop growth cycle, significantly improving prediction accuracy and real-time performance.

[0005] This application provides a yield prediction method based on crop growth models, including:

[0006] S101. Based on remote sensing data of the target crop's production area from the planting time to the current time, the full growth curve of the target crop is obtained using a pre-trained vegetation index growth prediction model.

[0007] S102, using a preset spatial heterogeneity analysis mechanism, generate the heterogeneity development curve and heterogeneity guidance curve of the target crop, perform differential analysis with the preset dual benchmark curves, and extract the target deviation range.

[0008] S103: Based on the target deviation interval, extract the NDVI multivariate feature vector of the full growth curve of the target crop, input it into the pre-trained crop yield prediction model, and output the predicted yield value of the target crop.

[0009] S104, Repeat steps S101 to S103 every preset period to update the predicted yield value of the target crop.

[0010] Preferably, the full growth curve of the target crop is used to represent the change of the NDVI value of the target crop over the entire growth cycle in a time series; the preset spatial heterogeneity analysis mechanism specifically includes:

[0011] S201, the production area is evenly divided into several sub-regions. Based on the average NDVI value of the pixels in each sub-region in each time unit, the spatial heterogeneity value of the production area in each time unit to date is obtained, and a heterogeneity development curve is generated.

[0012] S202, based on each time unit, filter the set of abnormal sub-regions, calculate the heterogeneous directional centroid coordinates of the set of abnormal sub-regions within the production area, and generate a heterogeneity directional curve based on the sequence of heterogeneous directional centroid coordinates of continuous time units relative to the heterogeneous direction values ​​of the production area.

[0013] Preferably, the spatial heterogeneity value within the time unit is calculated according to the following formula:

[0014]

[0015] in, Let n be the spatial heterogeneity value within the t-th time unit, and n be the total number of sub-regions of the production area. Let represent the NDVI value of the i-th sub-region in the t-th time unit. This represents the average NDVI value of all sub-regions within the production area during the t-th time unit.

[0016] Preferably, the set of abnormal sub-regions is represented as follows: , Let be the set of abnormal sub-regions in the t-th time unit. This represents the number of the i-th sub-region. The NDVI value of the i-th sub-region within the time unit. is the average NDVI value of all sub-regions within the production area in the time unit, and H is the spatial heterogeneity value within the time unit.

[0017] Preferably, the heterogeneous directional centroid coordinates of the abnormal sub-region set within the production area are obtained as follows:

[0018]

[0019] in, Let be the coordinates of the heterogeneous guided centroid in the t-th time unit. Let be the center coordinates of the j-th abnormal sub-region. Let be the DNVI residual value of the j-th anomalous sub-region. , Let represent the NDVI value of the j-th anomalous sub-region in the t-th time unit. This represents the average NDVI value of all sub-regions within the production area during the t-th time unit.

[0020] Preferably, in step S202, generating a heterogeneous guiding curve based on the heterogeneous directional value sequence of the heterogeneous centroid coordinates of the continuous time unit relative to the production area includes:

[0021] A1. Based on each time unit, the heterogeneous direction value is obtained according to the heterogeneous directional centroid coordinates of the abnormal subset and the center coordinates of the production area;

[0022] A2. The heterogeneous directional values ​​of continuous time units in the time series are combined into a heterogeneous directional value sequence. Using a pre-trained heterogeneity-guided prediction model, the heterogeneous directional value sequence of the complete growth cycle of the target crop in the production area is obtained. Based on the complete heterogeneous directional value sequence, curve fitting is performed to generate the heterogeneity-guided curve of the target crop.

[0023] Preferably, the preset dual reference curves include a first reference curve and a second reference curve, and are obtained as follows:

[0024] B1. Based on a pre-built historical database P= , For the k-th historical growth curve, and These are their corresponding heterogeneity development curves and heterogeneity guidance curves, respectively. is the absolute error between the historical predicted output value and the actual output value of the k-th historical growth curve;

[0025] B2. Calculate the DTW distance between the current full growth curve and the historical growth curves, and select the historical growth curve with the smallest absolute error value among all historical growth curves whose DTW distance value is less than the preset distance value as the reference curve.

[0026] B3. The heterogeneity development curve and heterogeneity guidance curve corresponding to the reference curve are respectively determined as the first reference curve and the second reference curve.

[0027] Preferably, in step S102, differential analysis is performed with the preset dual benchmark curves to extract the target deviation range, specifically including:

[0028] S301, based on dual benchmark curves, performs mutation detection to obtain the first mutation interval and the second mutation interval;

[0029] S302, compare the curve slope deviations of the first mutation interval on the first reference curve and the heterogeneity development curve, and the curve slope deviations of the second mutation interval on the second reference curve and the heterogeneity guidance curve, respectively, to determine the first deviation interval and the second deviation interval.

[0030] S303, take the intersection of the first deviation interval and the second deviation interval as the target deviation interval. , Let m be the m-th independent interval in the target deviation interval, where m represents the number of independent intervals.

[0031] Preferably, S103 specifically includes:

[0032] D1. Take the average NDVI of all curve segments on the full growth curve of the target crop corresponding to the target deviation interval, and form a deviation feature vector. , Let be the mean NDVI of the curve segment corresponding to the m-th independent interval on the full growth curve;

[0033] D2. Extract the NDVI peak point on the full growth curve of the target crop, calculate the first integral area value from the starting point to the peak point, the second integral area value from the peak point to the ending point, the maximum rate of change and the minimum rate of change of the growth curve, and form a key feature vector. , This is the area value of the first integral. This is the area value of the second integral. For the maximum rate of change, For the minimum rate of change, Peak value;

[0034] D3. Combine the deviation feature vector and the key feature vector to form the NDVI multivariate feature vector, input it into the pre-trained crop yield prediction model, and output the predicted yield value of the target crop.

[0035] Preferably, S302 specifically includes:

[0036] C1. Extract the first deviation interval as the interval where the slope deviation between the first reference curve and the heterogeneity development curve is greater than the first deviation threshold.

[0037] C2. Extract the second deviation interval as the interval where the slope deviation between the second reference curve and the heterogeneity guide curve is greater than the second deviation threshold.

[0038] The first mutation interval and the second mutation interval each include at least one independent interval.

[0039] This application also provides a yield prediction system based on a crop growth model, including: an acquisition module, a feature extraction module, and a prediction module;

[0040] The acquisition module is used to obtain the full growth curve of the target crop based on remote sensing data of the target crop's production area from the planting time to the current time, using a pre-trained vegetation index growth prediction model.

[0041] The feature extraction module is used to generate the heterogeneity development curve and heterogeneity guidance curve of the target crop using a preset spatial heterogeneity analysis mechanism, and to perform differential analysis with the preset dual benchmark curves to extract the target deviation interval; based on the target deviation interval, the NDVI multivariate feature vector of the full growth curve of the target crop is extracted.

[0042] The prediction module is used to input the NDVI multivariate feature vector into the pre-trained crop yield prediction model and output the predicted yield value of the target crop. The acquisition module, feature extraction module and prediction module are repeatedly executed every preset period to update the predicted yield value of the target crop.

[0043] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0044] By introducing a spatial heterogeneity analysis mechanism and a dual baseline curve generation method, this study effectively captures the spatial differentiation and environmental stress gradient during crop growth. Simultaneously, by utilizing differential analysis and target deviation interval extraction techniques, it focuses on the multivariate characteristics of NDVI, significantly improving the accuracy and real-time performance of crop yield prediction. Applicable to different crops and growing environments, it achieves accurate yield prediction at any point in the crop growth cycle, providing a strong scientific basis for agricultural management and decision-making, and helping to formulate countermeasures in advance.

[0045] The NDVI multivariate feature vector corresponding to the target deviation interval on the whole growth curve is extracted and input into the pre-trained crop yield prediction model to output the predicted yield value. This improves the targeting and accuracy of the training model. By using key features in the crop growth process to predict the final yield, the crop yield can be predicted more accurately, providing a scientific basis for agricultural management and decision-making. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the yield prediction method based on a crop growth model according to an embodiment of the present invention.

[0047] Figure 2 This is a structural block diagram of a crop growth model-based yield prediction system according to an embodiment of the present invention. Detailed Implementation

[0048] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0049] It should be noted that the terms "vertical," "horizontal," "up," "down," "left," "right," and similar expressions used in this article are for illustrative purposes only and do not represent the only possible implementation.

[0050] Unless otherwise defined, 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 invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0051] Example 1: Figure 1 This is a flowchart illustrating the yield prediction method based on a crop growth model according to an embodiment of the present invention.

[0052] like Figure 1 As shown, a yield prediction method based on a crop growth model includes the following steps:

[0053] S101. Based on remote sensing data of the target crop's production area from the planting time to the current time, a pre-trained vegetation index growth prediction model is used to obtain the full growth curve of the target crop. The full growth curve of the target crop is used to represent the change of the NDVI value of the target crop over the entire growth cycle in the time series.

[0054] Specifically, based on remote sensing data of the target crop production area from the planting date to the current day, the set of all pixels in the production area for each day is obtained. Based on the daily NDVI values ​​of all pixels, the daily average NDVI value of the production area (representing the overall vegetation growth status of the production area on that day) is obtained. A smooth NDVI (vegetation index) average curve on the time series is generated. A pre-trained vegetation index growth prediction model is used to predict the complete growth cycle curve of the target crop, and the full growth curve of the target crop is output.

[0055] The vegetation index growth prediction model can be based on an LSTM (Long Short Term Memory networks) prediction model, which is used to predict the complete NDVI curve based on a partial NDVI curve, that is, to predict the complete NDVI growth curve from the planting date to the maturity date (the end of the target crop growing season). The prediction method and training method can be referred to the description of relevant existing technologies, and this invention will not elaborate on them.

[0056] S102, using a preset spatial heterogeneity analysis mechanism, generates the heterogeneity development curve and heterogeneity guidance curve of the target crop, performs differential analysis with the preset dual benchmark curves, and extracts the target deviation range.

[0057] In some embodiments, the preset spatial heterogeneity analysis mechanism specifically includes:

[0058] S201 divides the production area into several sub-regions evenly. Based on the average NDVI value of each sub-region in each time unit, the spatial heterogeneity value of the production area in each time unit to date is obtained, and a heterogeneity development curve is generated.

[0059] The time unit can be set to daily or weekly, depending on the actual situation. Data is aggregated according to the time unit, and the spatial heterogeneity value of each time unit is calculated.

[0060] The heterogeneity development curve is set as the heterogeneity development curve of the complete growth cycle of the target crop. The complete cycle can be predicted using a pre-trained heterogeneity development prediction model, which will not be elaborated upon in this invention. It should be noted that, based on the spatial heterogeneity value sequence of continuous time units from the planting date to the current date, the spatial heterogeneity value sequence of the complete growth cycle of the target crop in the production area is obtained using the pre-trained heterogeneity development prediction model. Based on the complete spatial heterogeneity value sequence, curve fitting is performed to generate the heterogeneity development curve of the target crop.

[0061] Specifically, the spatial heterogeneity value of the production area within a time unit is calculated using the following formula:

[0062]

[0063] in, Let n be the spatial heterogeneity value within the t-th time unit, and n be the total number of sub-regions of the production area. Let represent the NDVI value of the i-th sub-region in the t-th time unit. This represents the average NDVI values ​​of all sub-regions within the production area at time unit t. Therefore, a larger spatial heterogeneity value indicates greater differences in the growth of the target crop within the production area, implying variations in soil parameters and water and fertilizer distribution, leading to differences in the internal and external growth environments of the entire production area.

[0064] It is understandable that the heterogeneity development curve of the target crop is obtained by curve fitting the spatial heterogeneity value sequence within a continuous time unit of the target crop's complete growth cycle.

[0065] Therefore, heterogeneity development curves are used to quantify the degree of uneven growth of target crops within production areas, capture environmental stress gradients (e.g., differences in soil fertility), reveal spatial differentiation dynamics, and identify sensitive periods for stress diffusion.

[0066] S202, based on each time unit, filter the set of abnormal sub-regions, calculate the heterogeneous directional centroid coordinates of the set of abnormal sub-regions within the production area, and generate a heterogeneity directional curve based on the sequence of heterogeneous directional centroid coordinates of continuous time units relative to the heterogeneous direction values ​​of the production area.

[0067] The set of abnormal subregions is represented as follows: , Let be the set of abnormal sub-regions in the t-th time unit. This represents the number of the i-th sub-region. The NDVI value of the i-th sub-region within the time unit. is the average NDVI value of all sub-regions within the production area in the time unit, and H is the spatial heterogeneity value within the time unit.

[0068] Specifically, the method for obtaining the heterogeneous oriented centroid coordinates of the abnormal sub-region set within the production area is as follows:

[0069]

[0070] in, Let be the coordinates of the heterogeneous guided centroid in the t-th time unit. ) represents the center coordinates of the j-th abnormal sub-region (a plane coordinate system can be constructed with the center of the production area as the origin, the horizontal direction as the x-axis, and the vertical direction as the y-axis to determine the center coordinates of the abnormal sub-region). Let be the DNVI residual value of the j-th anomalous sub-region. , Let represent the NDVI value of the j-th anomalous sub-region in the t-th time unit. Let be the average NDVI values ​​of all sub-regions within the production area in the t-th time unit. Based on this, the heterogeneous directional centroid coordinates of the set of anomalous sub-regions are calculated. The weight of each anomalous sub-region is the absolute value of its NDVI residual (i.e., the degree of deviation from the average). The larger the weight (i.e., the greater the degree of anomalousness), the greater the influence of that sub-region on the centroid position.

[0071] Specifically, based on the sequence of heterogeneous directional centroid coordinates relative to the heterogeneous direction values ​​of the production area in continuous time units, heterogeneous directional curves are generated, including:

[0072] A1. Based on each time unit, the heterogeneous direction value is obtained according to the heterogeneous directional centroid coordinates of the abnormal subset and the center coordinates of the production area.

[0073] The heterogeneous orientation value is calculated using the following formula:

[0074]

[0075] in, The heterogeneous direction value in the t-th time unit = ), g= ), where g is the center coordinate of the production area. Let be the ordinate of the heterogeneous guided centroid coordinates in the t-th time unit. Let x be the x-coordinate of the heterogeneous guided centroid coordinates in the t-th time unit.

[0076] A2. The heterogeneous directional values ​​of continuous time units in the time series are combined into a heterogeneous directional value sequence. Using a pre-trained heterogeneity-guided prediction model, the heterogeneous directional value sequence of the complete growth cycle of the target crop in the production area is obtained. Based on the complete heterogeneous directional value sequence, curve fitting is performed to generate the heterogeneity-guided curve of the target crop.

[0077] Therefore, the orientation angle of the heterogeneous guidance centroid relative to the center of the production area is calculated to generate the heterogeneity guidance curve. The change of the orientation angle can reflect the movement trend or improvement of the abnormal area (e.g., the spread of pests and diseases from a certain direction). The heterogeneity guidance curve is used to track the migration trajectory of the abnormal area, locate the direction of disaster spread, and reflect the growth characteristics of the target crop in the production area.

[0078] It should be noted that the pre-trained heterogeneity development prediction model and the pre-trained heterogeneity-oriented prediction model can employ LSTM-based prediction models. These models predict the spatial heterogeneity value sequence and heterogeneity direction value sequence within the complete growth cycle of the target crop, respectively, based on partial spatial heterogeneity value sequences and partial heterogeneity direction value sequences. The models are trained using a large number of historical growth curves and their corresponding spatial and directional heterogeneity values ​​to uncover the changing patterns over time, thereby obtaining the pre-trained heterogeneity development prediction model and the pre-trained heterogeneity-oriented prediction model. Specific training methods can be found in existing technologies, and will not be elaborated upon in this invention.

[0079] In some embodiments, the preset dual reference curves include a first reference curve and a second reference curve, used to establish a historical similar scenario reference system, and are obtained in the following way:

[0080] B1. Based on a pre-built historical database P= , For the k-th historical growth curve, and These are their corresponding heterogeneity development curves and heterogeneity guidance curves, respectively. denoted as the absolute error between the historical predicted output value and the actual output value of the k-th historical growth curve.

[0081] B2. Calculate the DTW distance between the current full growth curve and the historical growth curves (DTW distance is a dynamic programming algorithm used to measure the similarity between two time series, which will not be elaborated in this invention), and select the historical growth curve with the smallest absolute error value among all historical growth curves with DTW distance values ​​less than the preset distance value as the reference curve.

[0082] The preset distance value is determined based on historical data and expert experience and is used to measure the degree of difference between curves. The smaller the difference, the higher the similarity.

[0083] B3. The heterogeneity development curve and heterogeneity guidance curve corresponding to the reference curve are respectively determined as the first reference curve and the second reference curve.

[0084] In some embodiments, a difference analysis is performed with preset dual benchmark curves to extract target deviation intervals, which are used to focus on key phenological stages, specifically including:

[0085] S301, based on dual benchmark curves, performs mutation detection to obtain the first mutation interval and the second mutation interval.

[0086] Specifically, the first mutation region is obtained as follows:

[0087] The rate of change (which can also be understood as the slope value) of the first baseline curve is detected. If the direction of the rate of change changes (i.e., a positive-to-negative switch occurs), the mutation point (i.e. the time point when the sign of the first derivative changes) is marked. Each mutation point is used as the midpoint of the interval and is extended to both sides by a preset number of time units (e.g., set to 5) to obtain the mutation interval corresponding to each mutation point. The union of the mutation intervals of all mutation points is taken as the first mutation interval.

[0088] For example, for a mutation point t, taking the mutation point t as the midpoint of the interval, we obtain the corresponding mutation interval [t- t ], The preset number of time units allows the mutation interval to reflect the NDVI characteristics near the mutation point.

[0089] Specifically, the second mutation region is obtained as follows:

[0090] The rate of change of the second baseline curve is detected, and the intervals where the rate of change continues to rise and fall are taken as the second mutation intervals.

[0091] For example, the curve can be divided into bars, and the rate of change corresponding to each bar can be traversed in turn. When the rate of change is detected to be continuously rising or falling, the rate of change can be traversed continuously until the rate of change falls or rises, generating the corresponding mutation interval. Each mutation interval corresponds to a rate of change that is continuously rising or falling. The union of all mutation intervals is taken to obtain the second mutation interval.

[0092] S302, compare the slope deviations of the first mutation interval on the first reference curve and the heterogeneity development curve, and the slope deviations of the second mutation interval on the second reference curve and the heterogeneity guidance curve, respectively, to determine the first deviation interval and the second deviation interval.

[0093] Specifically, step S302 includes:

[0094] C1. Extract the first deviation interval as the interval (at least one) where the slope deviation between the first reference curve and the heterogeneity development curve is greater than the first deviation threshold.

[0095] For each independent interval (discontinuous interval) within the first abrupt change interval, calculate its slope on the first baseline curve and the heterogeneous development curve. Compare the absolute value of the difference between the two with a first deviation threshold. Independent intervals exceeding the first deviation threshold are designated as the first deviation interval. The first deviation threshold is determined based on historical and expert experience and measures the degree of difference in the changing trends of the first baseline curve and the heterogeneous development curve within the first abrupt change interval. If the difference in changing trends is too large, it indicates a significant difference between the two curves within this interval.

[0096] C2. Extract the second deviation interval as the interval (at least one) where the slope deviation of the curve on the second baseline curve and the heterogeneity guide curve is greater than the second deviation threshold.

[0097] For each independent interval (discontinuous interval) within the second mutation interval, calculate its slope on the second baseline curve and the heterogeneity-guided curve. Compare the absolute value of the difference between the two slopes with a second deviation threshold. Independent intervals exceeding this threshold are designated as the second deviation interval. The second deviation threshold, determined based on historical and expert experience, measures the degree of difference in the changing trends of the second baseline curve and the heterogeneity-guided curve within the second mutation interval. A significant difference indicates a substantial discrepancy between the two curves within this interval.

[0098] S303, take the intersection of the first deviation interval and the second deviation interval as the target deviation interval. , Let m be the m-th independent interval in the target deviation interval, where m represents the number of independent intervals, which is greater than or equal to 1.

[0099] It should be noted that in other embodiments, if there is no intersection, the curve slope deviation judgment threshold (first deviation threshold and second deviation threshold) in step S302 is appropriately adjusted until the first intersection occurs; in other embodiments, if the first mutation interval and the second mutation interval do not have an intersection, the independent interval with the largest curve slope deviation is directly taken as the first deviation interval and the second deviation interval, and the union is taken as the target deviation interval.

[0100] S103, based on the target deviation interval, extracts the NDVI multivariate feature vector of the full growth curve of the target crop, inputs it into the pre-trained crop yield prediction model, and outputs the predicted yield value of the target crop.

[0101] Specifically, step S103 includes:

[0102] D1. Take the average NDVI of all curve segments on the full growth curve of the target crop corresponding to the target deviation interval, and form a deviation feature vector. , denoted as the mean NDVI of the curve segment corresponding to the m-th independent interval on the full growth curve.

[0103] For example, the NDVI mean is calculated for all curve segments on the full growth curve corresponding to the target deviation interval, resulting in an NDVI value sequence, including the NDVI mean of at least one curve segment. It should be noted that the target deviation interval includes at least one independent interval, each independent interval corresponding to a curve segment and its NDVI mean. The NDVI value sequence is then used as the deviation feature vector.

[0104] D2. Extract the NDVI peak point from the full growth curve of the target crop (e.g., the physiological state corresponding to the heading stage), calculate the first integral area value from the starting point to the peak point, the second integral area value from the peak point to the ending point, the maximum rate of change and the minimum rate of change of the growth curve (which can be understood as the slope of the curve), and form the key feature vector. , This is the area value of the first integral. This is the area value of the second integral. For the maximum rate of change, For the minimum rate of change, This is the peak value.

[0105] The integral area of ​​the entire growth curve from the starting point to the peak point is:

[0106]

[0107] This represents the integral area of ​​the entire growth curve from the starting point to the peak point. Starting point The time point corresponding to the peak point. This is the full growth curve;

[0108] Similarly, the integral area from the peak point to the terminal point is:

[0109]

[0110] This represents the integral area of ​​the entire growth curve from the starting point to the peak point. As the endpoint, The time point corresponding to the peak point. This is the full growth curve.

[0111] D3. Combine the deviation feature vector and the key feature vector to form the NDVI multivariate feature vector, input it into the pre-trained crop yield prediction model, and output the predicted yield value of the target crop.

[0112] In some embodiments, the pre-trained crop yield prediction model is trained as follows:

[0113] E1. Collect a large number of historical full life cycle growth curves and their actual output values, and extract the NDVI multivariate feature vector for each historical full life cycle growth curve.

[0114] E2. Label the NDVI multivariate feature vectors using actual yield values ​​to form a training dataset. Use the training dataset to train a pre-selected neural network model, continuously optimize the model parameters, and obtain a crop yield prediction model.

[0115] It should be noted that when using NDVI multivariate feature vectors for training, different evaluation weight values ​​can be assigned to the bias feature vector and the key feature vector to differentiate their importance for yield prediction. During the training process, the evaluation weight values ​​of the two can be continuously adjusted to minimize the prediction error.

[0116] S104. Repeat steps S101 to S103 every preset period to update the predicted yield value of the target crop until the target crop is fully mature.

[0117] It should be noted that the predicted yield value for each cycle is synchronized to the production area management terminal, along with the full growth curve, heterogeneity development curve, and heterogeneity guidance curve information of the target crop in the production area. This information is used by the production area management terminal for timely scheduling and adjustment to optimize the growth of the target crop.

[0118] In summary, while the prediction models in related technologies are applicable to a variety of crops, they neglect the spatial heterogeneity of crop growth within the production area. This application improves the adaptability of the method to different growth environments and conditions and enhances its universality by introducing a spatial heterogeneity analysis mechanism. While achieving yield prediction at any point in the crop growth cycle, it also improves the accuracy and specificity of point-in-time prediction by focusing on key phenological stages through differential analysis and target deviation interval extraction.

[0119] By introducing a spatial heterogeneity analysis mechanism and a dual baseline curve generation method, this study effectively captures the spatial differentiation and environmental stress gradient during crop growth. Simultaneously, by utilizing differential analysis and target deviation interval extraction techniques, it focuses on the multivariate characteristics of NDVI, significantly improving the accuracy and real-time performance of crop yield prediction. Applicable to different crops and growing environments, it achieves accurate yield prediction at any point in the crop growth cycle, providing a strong scientific basis for agricultural management and decision-making, and helping to formulate countermeasures in advance to ensure food security.

[0120] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0121] The production area is divided into several sub-regions, and the spatial heterogeneity value within each time unit is calculated to generate a heterogeneity development curve. The set of abnormal sub-regions is screened, and the coordinates of the heterogeneity-oriented centroid are calculated to generate a heterogeneity-oriented curve. The degree of uneven crop growth within the production area is quantified, the environmental stress gradient is captured, and the spatial differentiation dynamics are revealed. The heterogeneity development curve and the heterogeneity-oriented curve can respectively reflect the spatial heterogeneity of crop growth within the production area and the migration trajectory of abnormal areas.

[0122] By selecting historical growth curves similar to the current full growth curve using DTW distance values, their corresponding heterogeneity development curves and heterogeneity guidance curves are determined as the first and second baseline curves, providing a reference standard for subsequent differential analysis. Through dual baseline curves, abnormal changes in the current crop growth process can be identified more accurately.

[0123] By comparing the slope deviations of the first and second baseline curves with the heterogeneity development curve and the heterogeneity guidance curve, the first deviation interval and the second deviation interval are determined, and their intersection is extracted as the target deviation interval. This allows us to focus on key phenological stages and identify key change intervals in the crop growth process. The target deviation interval can accurately reflect the abnormal change stages in the crop growth process, providing key information for subsequent yield prediction.

[0124] The NDVI multivariate feature vector corresponding to the target deviation interval on the whole growth curve is extracted and input into the pre-trained crop yield prediction model to output the predicted yield value. This improves the targeting and accuracy of the training model. By using key features in the crop growth process to predict the final yield, the crop yield can be predicted more accurately, providing a scientific basis for agricultural management and decision-making.

[0125] The above steps are repeated every preset period to update the predicted yield value of the target crop. As the crop growth cycle progresses, the yield prediction results are continuously optimized to improve the accuracy and real-time performance of yield prediction, providing dynamic support for agricultural management.

[0126] Furthermore, embodiments of the present invention also provide a yield prediction system based on crop growth models.

[0127] Figure 2 This is a structural block diagram of a crop growth model-based yield prediction system according to an embodiment of the present invention.

[0128] like Figure 2 As shown, the yield prediction system based on crop growth models includes: an acquisition module, a feature extraction module, and a prediction module.

[0129] Specifically, the acquisition module is used to obtain the full growth curve of the target crop based on remote sensing data of the target crop's production area from the planting time to the current time, using a pre-trained vegetation index growth prediction model; the feature extraction module is used to generate the heterogeneous development curve and heterogeneous guidance curve of the target crop using a preset spatial heterogeneity analysis mechanism, and perform differential analysis with the preset dual benchmark curves to extract the target deviation interval; based on the target deviation interval, the NDVI multivariate feature vector of the full growth curve of the target crop is extracted; the prediction module is used to input the NDVI multivariate feature vector into the pre-trained crop yield prediction model and output the predicted yield value of the target crop; the acquisition module, feature extraction module and prediction module are repeatedly executed every preset period to update the predicted yield value of the target crop.

[0130] It should be noted that other specific implementation details of the embodiments of the present invention can refer to the above-described method for yield prediction based on crop growth models.

[0131] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A yield prediction method based on crop growth models, characterized in that, include: S101. Based on remote sensing data of the target crop's production area from the planting time to the current time, the full growth curve of the target crop is obtained using a pre-trained vegetation index growth prediction model. S102, using a preset spatial heterogeneity analysis mechanism, generate the heterogeneity development curve and heterogeneity guidance curve of the target crop, perform differential analysis with the preset dual benchmark curves, and extract the target deviation range. The preset spatial heterogeneity analysis mechanism specifically includes: S201, uniformly dividing the production area into several sub-regions, obtaining the spatial heterogeneity value of the production area in each time unit based on the average NDVI value of each sub-region in each time unit, and generating a heterogeneity development curve; S202, based on each time unit, filtering the set of abnormal sub-regions, calculating the heterogeneous directional centroid coordinates of the abnormal sub-region set in the production area, and generating a heterogeneous directional curve based on the heterogeneous directional centroid coordinates of continuous time units relative to the heterogeneous direction value sequence of the production area; the preset dual reference curves include a first reference curve and a second reference curve, obtained by: B1, based on a pre-constructed historical database P= , For the k-th historical growth curve, and These are their corresponding heterogeneity development curves and heterogeneity guidance curves, respectively. B1. The absolute error between the historical predicted yield value and the actual yield value of the k-th historical growth curve; B2. Calculate the DTW distance between the current full growth curve and the historical growth curves, and select the historical growth curve with the smallest absolute error value among all historical growth curves whose DTW distance value is less than the preset distance value as the reference curve; B3. The heterogeneity development curve and heterogeneity guidance curve corresponding to the reference curve are respectively determined as the first reference curve and the second reference curve; S103, based on the target deviation interval, extract the NDVI multivariate feature vector of the entire growth curve of the target crop, input it into the pre-trained crop yield prediction model, and output the predicted yield value of the target crop; the NDVI multivariate feature vector includes a deviation feature vector and a key feature vector, specifically: take the NDVI mean of all curve segments on the entire growth curve of the target crop corresponding to the target deviation interval to form the deviation feature vector. , Let be the mean NDVI of the curve segment corresponding to the m-th independent interval on the whole growth curve; extract the NDVI peak point on the whole growth curve of the target crop, calculate the first integral area value from the starting point to the peak point, the second integral area value from the peak point to the ending point, the maximum rate of change and the minimum rate of change of the growth curve, and form the key feature vector. , This is the area value of the first integral. This is the area value of the second integral. For the maximum rate of change, For the minimum rate of change, Peak value; S104, Repeat steps S101 to S103 every preset period to update the predicted yield value of the target crop.

2. The yield prediction method based on crop growth models as described in claim 1, characterized in that, The full growth curve of the target crop is used to represent the change of the NDVI value of the target crop over time throughout its entire growth cycle.

3. The yield prediction method based on crop growth models as described in claim 2, characterized in that, The spatial heterogeneity value within the time unit is calculated according to the following formula: ； in, Let n be the spatial heterogeneity value within the t-th time unit, and n be the total number of sub-regions of the production area. Let represent the NDVI value of the i-th sub-region in the t-th time unit. This represents the average NDVI value of all sub-regions within the production area during the t-th time unit.

4. The yield prediction method based on crop growth models as described in claim 3, characterized in that, The set of abnormal sub-regions is represented as follows: , Let be the set of abnormal sub-regions in the t-th time unit. This represents the number of the i-th sub-region. The NDVI value of the i-th sub-region within the time unit. is the average NDVI value of all sub-regions within the production area in the time unit, and H is the spatial heterogeneity value within the time unit.

5. The yield prediction method based on crop growth models as described in claim 4, characterized in that, The method for obtaining the heterogeneous directional centroid coordinates of the set of abnormal sub-regions within the production area is as follows: ； in, Let be the coordinates of the heterogeneous guided centroid in the t-th time unit. () represents the center coordinates of the j-th abnormal sub-region. Let j be the NDVI residual value of the j-th anomalous sub-region. , Let represent the NDVI value of the j-th anomalous sub-region in the t-th time unit. This represents the average NDVI value of all sub-regions within the production area during the t-th time unit.

6. The yield prediction method based on crop growth models as described in claim 2, characterized in that, In step S202, a heterogeneous directional curve is generated based on the sequence of heterogeneous directional centroid coordinates of the continuous time unit relative to the heterogeneous direction value of the production area, including: A1. Based on each time unit, the heterogeneous direction value is obtained according to the heterogeneous directional centroid coordinates of the abnormal subset and the center coordinates of the production area; A2. The heterogeneous directional values ​​of continuous time units in the time series are combined into a heterogeneous directional value sequence. Using a pre-trained heterogeneity-guided prediction model, the heterogeneous directional value sequence of the complete growth cycle of the target crop in the production area is obtained. Based on the complete heterogeneous directional value sequence, curve fitting is performed to generate the heterogeneity-guided curve of the target crop.

7. The yield prediction method based on crop growth models as described in claim 2, characterized in that, In step S102, a difference analysis is performed with the preset dual benchmark curves to extract the target deviation range, specifically including: S301, based on dual benchmark curves, performs mutation detection to obtain the first mutation interval and the second mutation interval; S302, compare the curve slope deviations of the first mutation interval on the first reference curve and the heterogeneity development curve, and the curve slope deviations of the second mutation interval on the second reference curve and the heterogeneity guidance curve, respectively, to determine the first deviation interval and the second deviation interval. S303, take the intersection of the first deviation interval and the second deviation interval as the target deviation interval. , Let m be the m-th independent interval in the target deviation interval, where m represents the number of independent intervals.

8. A yield prediction system based on a crop growth model, applied to the yield prediction method based on a crop growth model as described in any one of claims 1 to 7, characterized in that, include: Acquisition module, feature extraction module, prediction module; The acquisition module is used to obtain the full growth curve of the target crop based on remote sensing data of the target crop's production area from the planting time to the current time, using a pre-trained vegetation index growth prediction model. The feature extraction module is used to generate the heterogeneity development curve and heterogeneity guidance curve of the target crop using a preset spatial heterogeneity analysis mechanism, and to perform differential analysis with the preset dual benchmark curves to extract the target deviation interval; based on the target deviation interval, the NDVI multivariate feature vector of the full growth curve of the target crop is extracted. The prediction module is used to input the NDVI multivariate feature vector into the pre-trained crop yield prediction model and output the predicted yield value of the target crop. The acquisition module, feature extraction module and prediction module are repeatedly executed every preset period to update the predicted yield value of the target crop.

Images