A method and system for evaluating the quality of rape planting based on big data analysis

Abstract

The application discloses a rapeseed planting quality evaluation method and system based on big data analysis, and relates to the technical field of big data analysis. The method comprises the following steps: collecting sample data of a target rapeseed planting environment based on a big data method, wherein the sample data comprises planting full record data and product quality data; constructing a mechanism-data coupling model according to the sample data, wherein the mechanism-data coupling model comprises a cascaded rapeseed mechanism model and a data-driven quality mapping model; acquiring a native management log of a to-be-evaluated land in a current planting period, and performing dynamic quality evaluation based on the native management log and the mechanism-data coupling model. The application effectively improves the accuracy of rapeseed planting quality evaluation.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, specifically to a method and system for evaluating the quality of rapeseed cultivation based on big data analytics. Background Technology

[0002] With the rapid development of smart agriculture and the deep application of big data technology in agricultural production, rapeseed planting quality assessment has gradually upgraded from the traditional post-harvest laboratory testing model to mid-harvest prediction. Existing technologies mostly use equipment such as drone remote sensing, weather stations, and IoT sensors to collect data such as vegetation indices and environmental factors, and use pure data-driven machine learning models to establish statistical correlations between data and rapeseed quality, thus partially realizing mid-harvest quality prediction.

[0003] However, existing technologies mostly rely on collecting data such as environment and vegetation indices and using machine learning to establish statistical correlations with quality. This purely data-driven model does not incorporate the intrinsic physiological mechanisms of rapeseed growth, resulting in insufficient model interpretability and low decision-making credibility. Furthermore, it is highly dependent on costly labeled data, which limits its widespread application. When faced with changes in environment, variety, or management measures, its extrapolation ability is weak and its prediction accuracy is insufficient. At the same time, it lacks dynamic tracking and early warning during the growth process, making it difficult to provide effective intervention guidance. Summary of the Invention

[0004] This invention provides a method and system for assessing rapeseed planting quality based on big data analysis, aiming to solve the technical problem of insufficient accuracy in existing rapeseed planting quality assessment technologies.

[0005] In view of the above problems, the present invention provides a method and system for evaluating the quality of rapeseed planting based on big data analysis.

[0006] In a first aspect, the present invention provides a method for evaluating the quality of rapeseed cultivation based on big data analysis, comprising: Sample data of the target rapeseed planting environment were collected based on big data methods, wherein the sample data included planting full record data and product quality data; Based on the sample data, a mechanism-data coupling model is constructed, wherein the mechanism-data coupling model includes a cascaded rapeseed mechanism model and a data-driven quality mapping model; Obtain the native management logs of the plot to be evaluated in the current planting cycle, and perform dynamic quality assessment based on the native management logs and the mechanism-data coupling model.

[0007] Secondly, the present invention provides a rapeseed planting quality assessment system based on big data analysis, comprising: The sample data acquisition module is used to collect sample data of the target rapeseed planting environment based on big data methods. The sample data includes planting full record data and product quality data. The coupling model construction module is used to construct a mechanism-data coupling model based on the sample data, wherein the mechanism-data coupling model includes a cascaded rapeseed mechanism model and a data-driven quality mapping model; The dynamic quality assessment module is used to obtain the native management logs of the plot to be assessed in the current planting cycle, and to perform dynamic quality assessment based on the native management logs and the mechanism-data coupling model.

[0008] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a method and system for assessing rapeseed planting quality based on big data analysis. It comprehensively collects rapeseed planting records and product quality sample data through big data analytics, providing solid data support for the assessment. Then, it constructs a mechanism-data coupling model that combines a cascaded rapeseed mechanism model and a data-driven quality mapping model, overcoming the shortcomings of pure data-driven models that lack support from growth mechanisms. Finally, it combines the original management logs of the plots to be assessed with the coupling model to conduct dynamic quality assessment, effectively improving the accuracy of rapeseed planting quality assessment, achieving dynamic monitoring of the entire rapeseed planting cycle, and providing a scientific basis for precision agricultural intervention. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart illustrating a method for assessing rapeseed planting quality based on big data analysis, provided in an embodiment of the present invention; Figure 2 A schematic diagram of a rapeseed planting quality assessment system based on big data analysis provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: Sample data acquisition module 11, coupled model construction module 12, dynamic quality assessment module 13. Detailed Implementation

[0011] This invention provides a method and system for assessing rapeseed planting quality based on big data analysis, which addresses the technical problem of insufficient accuracy in existing rapeseed planting quality assessment technologies.

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0013] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, this invention provides a method for evaluating the quality of rapeseed cultivation based on big data analysis, the method comprising: S100: Collect sample data of the target rapeseed planting environment based on big data methods, wherein the sample data includes planting full record data and product quality data.

[0015] In this embodiment of the invention, sample data of the target rapeseed planting environment is collected based on big data methods. The sample data includes complete planting records and product quality data. In big data-based crop quality assessment, the construction and performance of the model are highly dependent on the scale, quality, and representativeness of the sample data. If data is collected only for a single plot to be evaluated, problems such as data sparsity, incomplete periods, or missing historical records often arise, leading to insufficient model training and weak generalization ability. Therefore, this step aims to expand data sources and construct a standardized sample dataset covering multiple plots and multiple growth cycles, providing a sufficient, consistent, and reliable input foundation for the subsequent construction of a mechanism-data coupling model.

[0016] Step S100 in the method provided in this embodiment of the invention includes: Based on the intrinsic information of the land parcels to be evaluated, a set of parcels from the same source is determined; Based on the preset collection period constraints, the historical management logs of the land parcel to be evaluated and the set of parcels from the same source are extracted to obtain the sample data; The planting record data includes environmental data, agricultural management data, crop growth data, and corresponding product quality data from a preset historical planting cycle.

[0017] First, based on the intrinsic information of the plots to be evaluated, a set of source plots is determined. Intrinsic plot information refers to the inherent, long-term unchanging attributes of a plot, which is the key basis for judging the consistency of planting conditions. This includes geographical location (latitude and longitude range), soil type, crop varieties, climate zone, and altitude. A set of source plots refers to a collection of multiple plots with consistent or highly similar intrinsic information to the plots to be evaluated, whose planting conditions are comparable. Their historical data has direct reference value for model training. First, complete intrinsic information of the plots to be evaluated is extracted through an agricultural big data platform. Then, a matching threshold is set, and plots meeting the criteria are retrieved from the database, ultimately forming a set of source plots.

[0018] For example, the intrinsic information of the plot T001 to be evaluated is extracted: its geographical location is 105°-106°E, 30°-31°N, soil type is purple soil, the planted variety is Chuanyou 48, the climate zone is subtropical humid climate, and the altitude is 450-500 meters. A matching threshold of intrinsic information similarity ≥90% is set. Subsequently, the intrinsic information of all rapeseed planting plots in the regional agricultural big data platform is retrieved: 30 homologous plots T001-T030 are selected that share the same characteristics: planting Chuanyou 48, purple soil, subtropical humid climate zone, altitude 400-550 meters, and adjacent latitude and longitude ranges. All dimensions show a comprehensive similarity ≥90%. These 30 plots are then integrated to form a homologous plot set.

[0019] Secondly, based on the preset collection cycle constraints, historical management logs are extracted from the plots to be evaluated and the set of plots from the same source to obtain the sample data. The sample data includes complete planting record data and product quality data. The complete planting record data includes environmental data, agricultural management data, crop growth data, and corresponding product quality data from a preset historical planting cycle. The preset collection cycle constraint refers to a pre-defined time range for sample data collection, typically multiple complete planting cycles. The rapeseed planting cycle is approximately 8-9 months, ensuring data coverage of climate fluctuations and management differences across different years. Historical management logs refer to the raw operation and monitoring data recorded by the plots in past planting cycles, including raw records of environmental monitoring, agricultural operations, growth monitoring, and quality testing, without processing. Complete planting record data refers to a multi-dimensional data set covering the entire growth cycle of rapeseed from sowing to harvest, including environmental data, agricultural management data, and crop growth data. Product quality data refers to quality indicator data obtained from laboratory testing after rapeseed harvest, such as oil content, protein content, and erucic acid content. First, define the preset collection period, then traverse each plot in the same source plot group, extract the historical management logs of the corresponding period from the agricultural production management system and the IoT device storage module, and finally separate the planting full record data and product quality data from the logs, and organize them into standardized sample data in the format of plot number-planting period-data type-specific value.

[0020] For example, the preset collection period is the last 5 planting cycles, traversing T001 and the same source plot set T002-T030. Taking the planting cycle of plot T001 from September 2021 to May 2022 as an example: extract historical management logs: retrieve the complete logs of this cycle from the local agricultural production management platform, including daily environmental data, agricultural operation records, growth monitoring reports, and post-harvest quality test results. Breakdown of planting data: Environmental data: Average temperature on October 1, 2021: 18℃, cumulative sunshine hours: 6.5 hours; rainfall on October 3: 12mm; soil nitrogen content in mid-October: 25mg / kg; Agricultural management data: Sowing on September 28: 0.3kg / mu; Base fertilizer application on October 15: 10kg / mu pure nitrogen + 8kg / mu phosphorus and potassium fertilizer; Topdressing on November 22: 5kg / mu pure nitrogen; Spraying imidacloprid 30ml / mu on March 10, 2022; Crop growth data: NDVI values ​​monitored by drone every 10 days: 0.52 on November 10, 0.61 on November 20, and 0.73 on February 15, 2022. NDVI is the vegetation cover index, reflecting the degree of growth. Product quality data extraction: After harvest on May 10, 2022, laboratory tests showed an oil content of 42.3%, a protein content of 21.5%, and an erucic acid content of 0.8%. Sample data integration: The above data from 30 plots across 5 planting cycles were compiled into a standardized sample set, resulting in 150 complete cycle sample data sets (30 plots × 5 cycles). Each sample set includes complete planting records and corresponding product quality data.

[0021] In this embodiment of the invention, by setting a matching threshold to screen for plots of similar origin, the sample data is ensured to be highly matched with the planting background of the plots to be evaluated, while covering more microenvironments and management differences, thus improving the representativeness of the samples. The full planting record data covers all dimensions of environment, agricultural activities, growth, and quality, fully capturing the correlation between rapeseed growth and quality formation, providing comprehensive and accurate big data support for subsequent mechanism model calibration and quality mapping model training. The standardized sample collection and processing methods ensure uniform data format and clear labels, reducing the difficulty of subsequent data preprocessing and laying a solid foundation for the efficient construction and stable performance of the mechanism-data coupling model.

[0022] S200: Based on the sample data, construct a mechanism-data coupling model, wherein the mechanism-data coupling model includes a cascaded rapeseed mechanism model and a data-driven quality mapping model.

[0023] In this embodiment of the invention, a mechanism-data coupling model is constructed based on the sample data. This model includes a cascaded rapeseed mechanism model and a data-driven quality mapping model. The sample data is full-dimensional time-series data covering multiple plots and multiple planting cycles. Directly using this data for model training would result in data redundancy, unclear temporal correlations, and difficulty in capturing quality change patterns. In existing technologies, purely data-driven models often use complete cycle data for training, failing to focus on the correlation between growth status and quality changes at specific time periods, leading to poor model adaptability for dynamic assessment. Therefore, this step requires time-series segmentation, data preprocessing, and quality fluctuation quantification to transform the original sample data into input features and supervision signals that meet the model training requirements. Then, a cascaded structure of a rapeseed mechanism model and a quality mapping model is constructed, preserving the logic of the growth physiological mechanism while utilizing data-driven methods to improve prediction accuracy, providing model support for subsequent dynamic quality assessment.

[0024] Step S200 in the method provided in this embodiment of the invention includes: Define the sample segmentation window based on the preset quality assessment interval; Based on the sample segmentation window, data is cropped starting from the latest time node of each sample data item to obtain cropped sample data. The cropped sample data is traversed, the full planting record data is extracted and data preprocessing is performed to form full time-series sample data; Traverse the cropped sample data, extract the first product quality data and the second product quality data corresponding to the beginning and end of the planting full record data, calculate the quality data difference, and obtain the sample quality fluctuation data; Using the sample quality fluctuation data as supervision and the sample full-time series data as input, the mechanism-data coupling model is constructed and trained.

[0025] First, define the sample segmentation window based on the preset quality assessment interval. The preset quality assessment interval refers to the time frequency of subsequent dynamic assessments, i.e., predicting rapeseed quality at fixed intervals, which needs to match the frequency of growth monitoring in actual production, and is usually set to 10 days. The sample segmentation window is a time-series data segment defined to extract growth data and corresponding quality changes for a specific period, including the window time span and sliding step size, used to extract effective training segments from the complete cycle data. The window time span needs to cover a sufficient duration of growth changes, and the sliding step size is consistent with the quality assessment interval. First, determine the quality assessment interval based on the key growth period and monitoring frequency of rapeseed; then, set the time span and sliding step size of the sample segmentation window based on the assessment interval; finally, define the pruning rules according to the window parameters.

[0026] For example, the preset quality assessment interval is 10 days. Considering the significant quality changes during the rapeseed grain-filling period, the time span of the sample segmentation window is set to 30 days to cover the key period of quality formation, with a sliding step of 10 days, synchronized with the assessment interval. For any set of periodic samples, such as the 2021-2022 planting cycle of plot T001, sowing was on September 28, 2021, and harvesting was on May 10, 2022, totaling 225 days, the window cutting rule is: with the harvest date of May 10 as the final endpoint, the window is cut from back to front with a 30-day span + 10-day step, that is, each periodic sample can be divided into multiple non-overlapping time segments.

[0027] Secondly, based on the aforementioned sample segmentation window, data is pruned starting from the latest time node of each sample data item to obtain pruned sample data. A sample data item refers to the complete data set for a single plot and a single planting cycle, such as all daily data plus harvest quality data for plot T001 in the 2021-2022 cycle. The latest time node refers to the end point of the timeline for a specific sample data item, i.e., the rapeseed harvest date for that planting cycle, such as May 10, 2022 for plot T001. Data pruning refers to extracting a fixed span of time-series data forward from the harvest date (the end point of the window) according to the sample segmentation window rules, forming pruned samples focusing on the growth status of a specific period. All cycle samples are traversed, and the latest harvest date time node for each sample is determined one by one; using this node as the end point of the sample segmentation window, data is extracted forward according to the window's time span; pruning is repeated according to a sliding step size until the critical period for rapeseed quality formation is covered, such as from the grain-filling stage to harvest, approximately 60 days. Ultimately, multiple pruned samples are obtained for each cycle sample.

[0028] For example, taking the 2021-2022 periodic sample of plot T001 as an example, the latest harvest date is May 10, 2022, with a window span of 30 days and a step size of 10 days. The trimming process is as follows: First trimming: Using May 10 as the endpoint, 30 days of data from April 10 to May 10 are extracted to form trimmed sample S1; Second trimming: Sliding forward by 10 days, using April 30 as the endpoint, 30 days of data from March 31 to April 30 are extracted to form trimmed sample S2; Third trimming: Continuing to slide, using April 20 as the endpoint, 30 days of data from March 21 to April 20 are extracted to form trimmed sample S3; Finally, this periodic sample yields 3 trimmed samples, covering the critical 60 days of the grouting period. The remaining periodic samples are processed similarly, ultimately yielding 450 trimmed sample data from 150 periodic samples.

[0029] Further, the cropped sample data is traversed to extract the full planting record data and perform data preprocessing to form the full time-series sample data. Data preprocessing refers to a series of operations such as cleaning, aligning, interpolating, and standardizing the full planting record data in the cropped samples. The purpose is to eliminate data noise, fill in missing values, unify data dimensions, and ensure data usability. The full time-series sample data refers to multi-dimensional time-series data arranged in chronological order after preprocessing, containing key features of environment, agricultural activities, and growth, and serves as the input features for the model. All cropped samples are traversed, and the full planting record data is extracted one by one. First, data cleaning is performed to remove outliers, such as temperatures exceeding physiological thresholds and unreasonable NDVI values. Then, spatiotemporal alignment is performed, interpolating non-daily data, such as soil nitrogen content per ten-day period, into daily data to unify with the time axis of environment and growth data. Finally, standardization is performed to normalize all features to the [0,1] interval, eliminating the influence of dimensions, and ultimately forming the full time-series sample data arranged by day.

[0030] For example, taking the cropped sample S1 as an example: Data extraction: Extract the daily environmental data for this period, including temperature, sunshine, rainfall, and soil moisture; Agricultural data: No new agricultural operations were added, so the record is "none"; Growth data: NDVI values ​​for April 10, 20, and 30. Data cleaning: The average daily temperature of 42℃ recorded on April 18th exceeded the suitable growth temperature range of 5-25℃ for rapeseed and was identified as an outlier. It was replaced with the average temperature of 19℃ for the three days before and after. Spatiotemporal alignment: Soil nitrogen content was measured every ten days, with 23 mg / kg in mid-April and 21 mg / kg in late April. Linear interpolation was used to complete the daily data from April 10th to May 10th, such as 22.8 mg / kg on April 11th. The three NDVI values ​​were interpolated to the daily data, such as NDVI=0.69 on April 11th. Standardization: Features such as temperature (5-25℃) and NDVI (0.5-0.8) were normalized to the [0,1] interval, resulting in 30 daily data points, each containing 12 dimensions of features: 5 dimensions of environment + 2 dimensions of agricultural activities + 5 dimensions of growth. The remaining cropped sample data were processed in the same way, resulting in a set of 450 samples of full-time-series data.

[0031] Subsequently, the cropped sample data is traversed, and the first and second product quality data corresponding to the beginning and end of the planting record data are extracted. The difference in quality data is calculated to obtain sample quality fluctuation data. The first product quality data refers to the predicted final quality data corresponding to the start time point of the cropped sample window, that is, based on the growth data before the start time point, the quality indicators, such as oil content, are predicted at the final harvest of this cycle. The second product quality data refers to the measured quality indicators after the actual harvest of this planting cycle, that is, the actual quality data corresponding to the end point of the cropped sample window. The sample quality fluctuation data refers to the difference between the second product quality data and the first product quality data, which quantifies the degree of influence of the change in growth status within the corresponding period of the cropped sample on the final quality, and serves as a supervision signal for model training. All cropped samples are traversed. For each sample: the start time point of the window is determined, and the predicted final quality data corresponding to that time point is extracted to obtain the first product quality data; the measured quality data after the actual harvest of this cycle is extracted to obtain the second product quality data; the sample quality fluctuation data = first product quality data - second product quality data, including positive / negative / zero values, corresponding to quality improvement, decline, and no change, respectively.

[0032] For example, taking the cut sample S1 from plot T001 for the 2021-2022 period as an example: First product quality data is extracted: the window starts on April 10th. Based on the growth data of this plot from September 28th, 2021 to April 9th, 2022, the final oil content is predicted to be 41.1% using a linear regression model, i.e., first product quality data = 41.1%; Second product quality data is extracted: the actual measured oil content after harvest in this period is 42.3%, i.e., second product quality data = 42.3%; Quality fluctuation data = 42.3% - 41.1% = 1.2%, meaning the quality fluctuation data of this cut sample is +1.2%, indicating that the growth status was optimized within these 30 days, resulting in a 1.2% increase in the final oil content. Similarly, the quality fluctuation data for cut sample S2 is +1.8%; the quality fluctuation data for cut sample S3 is +0.9%. A total of 450 sample quality fluctuation data are obtained from the 450 cut samples.

[0033] Finally, using the sample quality fluctuation data as supervision and the sample full-time series data as input, the mechanism-data coupling model is constructed and trained.

[0034] The mechanism-data coupling model is constructed and trained using the sample quality fluctuation data as supervision and the sample full-time series data as input, including: Based on the sample crop growth time series data in the sample full-time series data, extract the training set of latent state variables; Based on the prior mechanism model, the rapeseed mechanism model is initialized and constructed. The rapeseed mechanism model is calibrated using the training set of the hidden state variables and the full-time data of the samples.

[0035] The implicit state variables include at least: nitrogen stress factor, water stress factor, heat stress factor, and grain allocation coefficient.

[0036] First, a training set of latent state variables is extracted from the sample crop growth time-series data in the full-time data of the samples. These latent state variables include at least: nitrogen stress factor, water stress factor, heat stress factor, and grain allocation coefficient. The sample crop growth time-series data refers to the observable time-series data reflecting the growth status of rapeseed in the full-time data of the samples. It serves as an indirect basis for characterizing the crop's physiological state and includes NDVI (Vegetation Cover Index), LAI (Leaf Area Index), and biomass estimates obtained from UAV / satellite remote sensing, arranged daily to form a time-series sequence. The training set of latent state variables refers to the set of indirect physiological state variables strongly correlated with growth, derived from the sample crop growth time-series data and combined with the physiological laws of rapeseed growth. This set is used for subsequent calibration and verification of the mechanistic model. Each cropped sample corresponds to a set of daily latent state variable sequences. All cropped samples were traversed, and crop growth time-series data such as NDVI and LAI were extracted for each sample. Rapeseed physiological knowledge was used, such as nitrogen stress leading to a decrease in NDVI and a slow increase in LAI, and an increase in the grain allocation coefficient accompanied by the transformation from vegetative growth to reproductive growth. Based on this rapeseed physiological knowledge, association rules between growth data and four types of latent state variables were established. Through statistical inversion such as multiple linear regression and gradient boosting regression, the values ​​of the four latent state variables were inferred from each daily growth data, ultimately forming a training set of 30 daily data points for each cropped sample, with each data point containing four-dimensional variables.

[0037] For example, taking the cropping sample S1 of plot T001 in the 2021-2022 cycle as an example: extracting crop growth time series data: the crop growth time series data of this sample is: daily NDVI value 0.68-0.75, LAI value 3.2-4.1, a total of 2-dimensional daily time series data. Association rules were established: Based on the physiological characteristics of the Sichuan Oil 48 variety, the following predefined rules were established: When the normalized vegetation index (NDI) is greater than or equal to 0.72, the corresponding nitrogen stress factor is less than or equal to 0.1, and this value range corresponds to a stress-free state; when the NDI is greater than or equal to 0.65 and less than 0.72, the corresponding nitrogen stress factor is between 0.1 and 0.5, and this value range corresponds to a mild stress state; when the leaf area index (LAI) growth rate is greater than or equal to 0.03 / day, the corresponding grain allocation coefficient is less than or equal to 0.5, and this value range corresponds to the vegetative growth-dominated stage; when the LAI growth rate is less than 0.03 / day, the corresponding grain allocation coefficient is greater than 0.5, and this value range corresponds to the reproductive growth-dominated stage. Latent state variables were inverted: A gradient boosting regression model was used, with the NDI and LAI as input features, to invert the specific values ​​of the four latent state variables each day. For example, on April 20th: the Normalized Difference Vegetation Index (NDVI) was 0.73, the Leaf Area Index (LAI) was 3.8, and the model back-calculated nitrogen stress factor was 0.08, water stress factor was 0.05, heat stress factor was 0.03, and the grain allocation coefficient was 0.62. Training set formation: This cropped sample contains 30 daily data points, each corresponding to 4 latent state variables, resulting in 30 × 4 = 120 variable values. These are then integrated to form the latent state variable training set for this sample. A total of 450 cropped samples form 450 latent state variable training sets, with a total of 450 × 30 × 4 = 54,000 variable values.

[0038] Secondly, based on the prior mechanism model, the rapeseed mechanism model is initialized and constructed. The prior mechanism model refers to a mature model recognized in the agricultural field, built upon the physiological processes of rapeseed growth, possessing a general growth simulation framework. It can be adapted to specific varieties and environments through parameter adjustments, such as APSIM-Canola and DSSAT-Canola models. Rapeseed mechanism model initialization refers to building a mechanism model framework adapted to this scheme by loading general parameters and basic attribute parameters of the target variety based on the prior mechanism model, enabling it to possess preliminary growth simulation capabilities. A mature prior mechanism model in the agricultural field, such as APSIM-Canola, is selected as the basic framework; publicly available physiological parameters of the target variety are collected, such as specific leaf area, light energy utilization efficiency, upper limit of grain filling rate, and critical temperature for oil synthesis, as initial variety parameters for the model; general environmental response parameters are loaded, such as the suitable growth temperature range of 5-25℃ and nitrogen absorption efficiency coefficient; inputs include environmental data and agricultural management data, and outputs include implicit state variables and simulated growth values. The interface format for model input and output is defined to complete the initial construction of the rapeseed mechanism model.

[0039] For example, the APSIM-Canola model was selected as the prior mechanistic model, which already covers the simulation module of the complete physiological process of rapeseed from sowing to harvest. Publicly available physiological parameters of rapeseed variety Chuanyou 48 were collected, and the initial parameters of the model were set as follows: specific leaf area 28. / kg, light energy utilization efficiency 2.8g / MJ, maximum grain filling rate 0.025 The critical temperature for oil synthesis is 10-22℃, and the saturation concentration for nitrogen uptake is 35mg / kg. General environmental response parameters are set: suitable growth temperature 5-25℃, exceeding this range triggers heat / cold stress; suitable soil moisture range 15%-25%, exceeding this range triggers water stress. The model input is set as daily environmental data, including temperature, sunshine duration, rainfall, soil nitrogen content, soil moisture, and agricultural management data, including fertilization time / amount and irrigation time / amount. The output consists of daily implicit state variables, including nitrogen stress factor, water stress factor, heat stress factor, grain allocation coefficient, and simulated growth values ​​NDVI and LAI. The interface formats for model input and output are defined: the input interface uses JSON data transmission format, supporting daily batch submissions; the output interface uses JSON data transmission format, returning simulation results daily, completing the initialization of the rapeseed mechanistic model.

[0040] Furthermore, the rapeseed mechanism model is calibrated using the training set of the hidden state variables and the full-time data of the samples.

[0041] The process of calibrating the rapeseed mechanism model using the latent state variable training set and the full-time data of the samples includes: Based on the full-time-series data of the sample, extract the time-series data of the sample environment and the time-series data of the sample agricultural management. The regulation response time of the target rapeseed variety is obtained by combining prior knowledge, and the time series data of the sample environment, the time series data of the sample agricultural management, and the training set of the latent state variables are time series realigned according to the regulation response time. Based on the temporal realignment results, the sample environmental temporal data and the sample agricultural management temporal data, the rapeseed mechanism model is driven to perform verification simulation to obtain the implicit state variable verification set. The rapeseed mechanism model is calibrated by adjusting the model parameters with the goal of minimizing the residual between the latent state variable validation set and the latent state variable training set.

[0042] First, based on the full-time series data of the samples, environmental time-series data and agricultural management time-series data are extracted. Agricultural management time-series data refers to the daily time-series data recording human intervention operations within the full-time series data, directly affecting crop growth, including fertilization, irrigation, and pesticide application. All cropped samples are traversed, and environmental and agricultural operation-related features are extracted from the 30 daily full-time series data points of each sample. These features are then organized in the format of daily timestamp + data type + specific numerical value to form the environmental and agricultural management time-series data corresponding to each sample.

[0043] For example, taking sample S1 of plot T001 from the 2021-2022 period as an example: the full-time series data of this sample is the data after preprocessing S200, containing 12-dimensional daily features. Five-dimensional environmental data are extracted from the 12-dimensional features and organized into daily sequences, for example: April 20: average daily temperature 19℃, cumulative sunshine hours 7.2 hours, daily rainfall 0mm, soil nitrogen content 21mg / kg, soil moisture 18%; April 21: average daily temperature 20℃, cumulative sunshine hours 6.8 hours, daily rainfall 5mm, soil nitrogen content 20.8mg / kg, soil moisture 20%. Two-dimensional agricultural data are extracted from the 12-dimensional features. There were no new agricultural operations during the corresponding time period for this sample; the daily records are: fertilization operation: none, irrigation operation: none; the topdressing records during the budding stage of the corresponding planting cycle are simultaneously included in this dataset for subsequent response time calculations. Similarly, time-series data of the sample environment and time-series data of sample agricultural management were obtained from 450 cropped samples.

[0044] Secondly, the regulatory response time of the target rapeseed variety is obtained by combining prior knowledge, and the time series data of the sample environment, the time series data of the sample agricultural management, and the training set of the latent state variables are time-series realigned according to the regulatory response time. The regulatory response time refers to the delay time after the target rapeseed variety is subjected to environmental changes or agricultural operations, resulting in a significant change in its physiological state. Determined based on prior knowledge of rapeseed physiology, it is a key parameter to ensure the timing match between input and output. Time series realignment refers to adjusting the starting point of the time axis of the sample environmental time series data and the sample agricultural management time series data according to the regulatory response time, so that the regulatory event and the corresponding physiological state change correspond precisely in time, eliminating causal misalignment caused by response delay. Based on prior knowledge in the agricultural field, such as variety cultivation manuals and physiological research literature, the response time of the target variety to different regulations is determined, such as the response time of fertilization affecting nitrogen absorption and the response time of irrigation affecting water stress relief. All cropped samples are traversed to extract the occurrence time of key regulatory events in each sample. The time axis of the sample environmental time series data and the sample agricultural management time series data are adjusted synchronously with the occurrence time of the regulatory event plus the response time as the new time starting point. The time axis of the training set of hidden state variables is aligned with the time axis of the adjusted input data to complete the time realignment of the three.

[0045] For example, the regulation response time is obtained as follows: Based on the prior knowledge of cultivation of Sichuan Oil 48, the key regulation response times are determined as follows: the absorption response time of topdressing (nitrogen) during the budding stage is 7 days; the response time of soil moisture stress relief after irrigation is 3 days; and the response time of heat stress triggered by high temperature (>25℃) is 1 day. Regulation events are extracted: Taking the 2021-2022 cycle of plot T001 corresponding to the cut sample S1 as an example, the key regulation event is topdressing (pure nitrogen 5 kg / mu) during the budding stage on November 22, 2021. Time series adjustment: Taking the topdressing occurrence time (November 22) + response time (7 days) = November 29 as the new starting point, the time axis of the sample environmental time series data and the sample agricultural management time series data is extracted from November 29, corresponding to the period from April 10 to May 10 of the cut sample S1. Realign the training set of latent state variables: Fully align the training set of latent state variables (April 10 - May 10) of the cropped sample S1 with the adjusted input data timeline to ensure that the environmental / agricultural data after November 29 correspond causally with the physiological state changes after April 10.

[0046] Furthermore, based on the time-series realignment results, the sample environmental time-series data, and the sample agricultural management time-series data, the rapeseed mechanism model is driven to perform validation simulations to obtain a validation set of latent state variables. Validation simulation refers to inputting the time-series realigned input data into the initialized, uncorrected rapeseed mechanism model, allowing the model to simulate the growth process according to physiological mechanisms, and outputting simulated latent state variables for comparison and validation with the training set. The latent state variable validation set refers to the set of simulation results output by the rapeseed mechanism simulation that has the same dimensions as the latent state variable training set, and serves as the basis for evaluating the initial accuracy of the rapeseed mechanism model. All the realigned sample environmental time-series data and sample agricultural management time-series data are input into the initialized rapeseed mechanism model one by one in daily order; simulation parameters are set, such as a simulation step size of 1 day, covering the 30-day period of the cropped samples; the rapeseed mechanism model is run, and the four latent state variables for each day are calculated through the internal physiological module; the simulation results of all samples are organized by sample number-day-variable type to form a latent state variable validation set.

[0047] For example, input realigned data: The daily environmental data and agricultural management data after realigning the cropped sample S1 are input into the initial rapeseed mechanism model on a daily basis; simulation parameters are set: the simulation step size is 1 day, the simulation period is 30 days (consistent with the cropped sample S1), and the initial rapeseed mechanism model runs according to the initial physiological parameters of Sichuan rapeseed 48; the initial rapeseed mechanism model outputs simulated values ​​of four latent state variables daily by simulating processes such as nitrogen absorption and photosynthetic product allocation. For example, on April 20th: the simulated nitrogen stress factor is 0.12, water stress factor is 0.07, heat stress factor is 0.04, and grain allocation coefficient is 0.58. The 30 daily simulation results of this cropped sample S1 constitute the validation set of the latent state variables for this sample; 450 cropped samples form 450 sets of latent state variable validation sets, totaling 450 × 30 × 4 = 54,000 variable values.

[0048] Finally, the rapeseed mechanism model is calibrated by adjusting model parameters to minimize the residuals between the latent state variable validation set and the latent state variable training set. Residuals, such as absolute error and mean squared error (MSE), are the differences between the simulated values ​​of the latent state variable validation set and the inversely calculated values ​​of the latent state variable training set; they are core indicators for measuring the model's simulation accuracy. Parameter adjustment involves using optimization algorithms to adjust key model parameters, such as variety physiological parameters and environmental response coefficients, to reduce residuals and make the model simulation results closer to actual physiological states. The process includes calculating the overall residuals between the latent state variable validation set and the training set for all cropped samples, such as the mean squared error (MSE) of the four variables; screening key adjustable parameters to exclude fixed physiological constants, such as the rapeseed basal metabolic coefficient; using parameter optimization algorithms, such as particle swarm optimization, to adjust key parameters within a reasonable range for the target species' physiological parameters; iteratively adjusting parameters and rerunning the model, calculating the residuals after each adjustment; and determining the fit between the simulated values ​​of the four variables and the training set when the residuals reach a preset minimum threshold, such as an average MSE ≤ 0.001. Stop adjusting when the time is up, and complete the model calibration.

[0049] For example, the initial residuals are calculated as follows: Comparing the validation set and training set with 450 samples, the initial residuals are: nitrogen stress factor MSE = 0.003, water stress factor MSE = 0.002, heat stress factor MSE = 0.001, grain allocation coefficient MSE = 0.004, and the mean MSE = 0.0025. The average residuals for the four variables are... =0.78, below the threshold. Determine adjustable parameters: Screen key parameters related to Chuanyou 48: initial specific leaf area 28 Initial nitrogen uptake efficiency coefficient: 0.8; initial soil moisture suitable threshold: 15%-25%; initial critical temperature for heat stress: 25℃. Iterative parameter adjustment: Using particle swarm optimization algorithm, parameters were adjusted within a reasonable range: specific leaf area was reduced to 26. The nitrogen uptake efficiency coefficient was increased to 0.92; the suitable soil moisture threshold was adjusted to 16%-24%; and the critical temperature for heat stress was increased to 26℃. Verification of the correction effect: The initial rapeseed mechanism model was rerun, and the residuals of 450 samples were calculated: the mean MSE = 0.0009, and the mean values ​​of the four variables were... =0.89, which satisfies the preset threshold MSE≤0.001. If the value is ≥0.85, stop adjusting and complete the calibration of the rapeseed mechanism model.

[0050] The mechanism-data coupling model, constructed and trained using the sample quality fluctuation data as supervision and the sample full-time series data as input, further includes: The full-time data of the samples and the training set of the latent state variables are time-series aligned and dimension-concatenated to obtain the sample fusion feature vector; A quality mapping model based on regression analysis is constructed, and the quality mapping model is iteratively trained using the sample fusion feature vector as training input and the sample quality fluctuation data as supervision. Using the quality mapping model as the downstream model and the rapeseed mechanism model as the upstream model, the models are cascaded to generate the mechanism-data coupling model.

[0051] First, the full-time series data of the samples and the training set of the latent state variables are time-series aligned and dimensionally concatenated to obtain the sample fusion feature vector. Time-series alignment refers to re-verifying and matching the timelines of the full-time series data of the samples and the training set of the latent state variables to ensure that the daily timestamps of the two data sequences correspond perfectly, eliminating minor time deviations caused by data processing procedures. Dimensional concatenation refers to merging the feature dimensions of the full-time series data of the samples at the same timestamp with those of the training set of the latent state variables to form a comprehensive feature sequence that includes both external growth characteristics and internal physiological characteristics. The sample fusion feature vector is the daily comprehensive feature set formed after time-series alignment and dimensional concatenation of each cropped sample, and it serves as the input feature of the quality mapping model. Iterate through all cropped samples, and for each sample, extract its corresponding full-time data and latent state variable training set; using the daily timestamp as a benchmark, verify the consistency of the time axis of the two data sequences to ensure that the 30 daily data points correspond one-to-one; for each daily data point, concatenate the 12-dimensional features of the full-time data and the 4-dimensional features of the latent state variable training set in sequence; arrange the concatenated features of the 30 daily data points in chronological order to form the sample fusion feature vector for each sample.

[0052] For example, taking the 2021-2022 periodic cropped sample S1 of plot T001 as an example: Time alignment: Extract the full time-series data and the training set of latent state variables for this sample, verifying that both timestamps are from April 10th to May 10th, with no time deviation. Dimensional concatenation: Concatenate the single data point from April 20th, merging the 12-dimensional full time-series features and the 4-dimensional latent state variables sequentially to form a comprehensive feature of 16 dimensions per day. Forming a fusion vector: Arrange the 30 daily 16-dimensional features of this sample in chronological order to form a 30×16 fusion feature vector, covering both external growth state and internal physiological characteristics.

[0053] Secondly, a quality mapping model based on regression analysis is constructed, and the fused sample feature vector is used as the training input, while the sample quality fluctuation data is used as supervision to iteratively train the quality mapping model. The quality mapping model is a data-driven model built based on regression analysis algorithms. Its function is to establish a mapping relationship between the fused sample feature vector and the sample quality fluctuation data, and output the predicted value of the quality fluctuation. Iterative training refers to repeatedly adjusting the model hyperparameters and updating the model weights to gradually reduce the deviation between the predicted value output by the quality mapping model and the sample quality fluctuation data until a preset accuracy threshold is reached. A quality mapping model is constructed using regression algorithms adapted to temporal features, such as XGBoost regression and random forest regression. The sample fusion feature vectors of all samples are used as the input dataset, and the corresponding sample quality fluctuation data are used as the supervision dataset, divided into training and test sets in a 7:3 ratio. Initial values ​​for model hyperparameters, such as decision tree depth and learning rate, are set. The training set is input into the model for initial training, and the predicted quality fluctuation values ​​are output. The error between the predicted values ​​and the training and supervision data, such as the root mean square error (RMSE), is calculated, and the hyperparameters are adjusted using a grid search method. This iterative process of training, error calculation, and parameter adjustment is repeated, while the model's generalization ability is verified using the test set, until the error reaches a preset threshold of RMSE ≤ 0.3%. The optimal model parameters are saved, completing the training of the quality mapping model.

[0054] For example, the XGBoost regression algorithm is selected to construct a quality mapping model, adapting to the mapping requirements of 16-dimensional time-series fusion features and single-value quality fluctuation data. Dataset partitioning: Of the 450 samples, 315 are used as the training set and 135 as the test set. Initial training and error calculation: Initial hyperparameters are set as follows: decision tree depth 6, learning rate 0.1, and number of iterations 100. The training set RMSE is calculated to be 0.45% for the training set and 0.52% for the test set, below the threshold of RMSE ≤ 0.3%. Hyperparameter adjustment and iteration: The hyperparameters are adjusted using grid search to: decision tree depth 8, learning rate 0.05, and number of iterations 150, and retrained. Training completion: The final training set RMSE is 0.22%, and the test set RMSE is 0.28%, meeting the preset thresholds. The model parameters are saved, and the quality mapping model training is complete.

[0055] Furthermore, using the quality mapping model as the downstream model and the rapeseed mechanism model as the upstream model, a model cascade is performed to generate the mechanism-data coupling model. Model cascading refers to connecting the rapeseed mechanism model and the quality mapping model in series according to data flow logic, with the output of the upstream model serving as part of the input of the downstream model, forming a complete link of input-intermediate output-final output. The upstream model, i.e., the rapeseed mechanism model, receives environmental and agricultural data and outputs implicit state variables, i.e., intrinsic physiological characteristics, providing mechanistic support for the downstream model. The downstream model, i.e., the quality mapping model, receives the fusion features of full-time data of the sample and the implicit state variables output by the upstream model, outputs predicted values ​​of quality fluctuations, and achieves accurate mapping from growth data to quality changes. Define the data flow interface between the upstream and downstream models to ensure that the format of the implicit state variables output by the upstream rapeseed mechanism model is consistent with the feature format required by the downstream quality mapping model. Set up the cascading logic: input environmental time series data and agricultural management time series data into the upstream rapeseed mechanism model, and the model outputs implicit state variables. Concatenate the dimensions of the implicit state variables with the full time series data of the samples to form a fused feature vector. Input the fused feature vector into the downstream quality mapping model, and the model outputs the predicted value of quality fluctuation. Verify the smoothness and prediction accuracy of the cascading link. If the requirements are met, fix the cascading relationship and generate the mechanism-data coupling model.

[0056] For example, the interface definition unifies the output format of the upstream rapeseed mechanism model and matches the input feature format of the downstream quality mapping model. Cascade logic construction: Input daily environmental data and agricultural management data from plot T001 for the 2023-2024 period to the calibrated rapeseed mechanism model. The rapeseed mechanism model outputs the daily latent state variables for this period, such as nitrogen stress factor 0.07-0.10 and water stress factor 0.04-0.06. This latent state variable (4-dimensional) is concatenated with the full-time series data (12-dimensional) of the same period samples to form a 16-dimensional fused feature vector. The fused feature vector is input into the trained quality mapping model, which outputs the predicted quality fluctuation value for this period, such as oil content fluctuation +1.1%. Cascade validation: Select periodic data from 10 plots that did not participate in the training to test the prediction accuracy of the cascade model. RMSE=0.29%, meeting the requirements. Fix the cascade relationship and generate a complete mechanism-data coupling model.

[0057] In this embodiment of the invention, a mechanism-data coupling model is constructed through a complete process: sample time-series pruning, mechanism model construction and correction, feature fusion, mapping model training, and model cascading. First, the full-cycle sample data is pruned and preprocessed based on a preset segmentation window to form full-time-series sample data focusing on the critical period of quality formation. Quality fluctuation data is extracted as a model supervision signal, effectively improving data utilization efficiency and relevance. Then, the training set of latent state variables is inferred from growth data, and the initialization and correction of the rapeseed mechanism model are completed by combining environmental and agricultural data. This solves the problem of poor parameter adaptability in general mechanism models, improves the simulation accuracy of the rapeseed's internal physiological state, and provides the model with solid physiological logic support. Finally, through time-series alignment and dimensional concatenation, the external parameters of the full-time-series sample data are integrated. By combining the intrinsic physiological characteristics of growth features and implicit state variables, a more comprehensive fusion feature vector is formed, which makes up for the shortcomings of single data dimensions. On this basis, the quality mapping model trained accurately captures the correlation between growth, physiology and quality fluctuations. Finally, the mechanism-data coupling model formed by the cascading of upstream and downstream models combines the interpretability of the upstream mechanism model with the high-precision prediction capability of the downstream data-driven model. It overcomes the shortcomings of insufficient prediction accuracy of pure mechanism models and solves the problem of weak generalization ability of pure data-driven models, providing a core technology carrier for the subsequent dynamic evaluation of rapeseed planting quality.

[0058] S300: Obtain the native management log of the plot to be evaluated in the current planting cycle, and perform dynamic quality evaluation based on the native management log and the mechanism-data coupling model.

[0059] In this embodiment of the invention, the native management log of the plot to be evaluated during the current planting cycle is obtained, and dynamic quality assessment is performed based on the native management log and the mechanism-data coupling model. The quality status of the plot to be evaluated during the current planting cycle is in a dynamic state of change. Relying solely on static detection after harvest cannot timely grasp the fluctuation patterns during the quality formation process, nor can it provide real-time basis for optimizing field management. Existing static assessment methods suffer from problems such as lag and inability to trace the causes of quality changes. However, based on the previously constructed mechanism-data coupling model, combined with real-time data and native management data of the current cycle, dynamic segmented assessment can accurately capture the quality fluctuation trends at different growth stages, achieving process-oriented, traceable, and intervention-oriented quality assessment, thus overcoming the limitations of static assessment.

[0060] Step S300 in the method provided in this embodiment of the invention includes: Obtain the native management log of the plot to be evaluated in the current planting cycle, and extract the corresponding native management time series data; The real-time external source record information of the plot to be evaluated is obtained and time-series data is generated to obtain real-time environmental time-series data and real-time crop growth time-series data. Time-series alignment is performed on the native management time-series data, the real-time environment time-series data, and the real-time crop growth time-series data. The timing alignment results are segmented according to the preset quality assessment interval to obtain multiple segmented timing data with temporal sequence relationship; According to the aforementioned temporal sequence, multiple segmented time series data are sequentially input into the mechanism-data coupling model for segmented quality assessment to obtain a predicted quality fluctuation dataset; By combining preset typical product quality data with the quality fluctuation dataset, dynamic quality assessment results are obtained.

[0061] First, the original management logs of the plot to be evaluated during the current planting cycle are obtained, and the corresponding original management time-series data is extracted. The original management logs refer to the raw agricultural operation logs recorded by growers and the agricultural management platform during the current planting cycle of the plot to be evaluated. These logs are unprocessed and contain information on the entire process of sowing, fertilizing, irrigating, and applying pesticides. The original management time-series data refers to the time-series agricultural data extracted from the original management logs and organized by daily timestamps. The format is consistent with the agricultural management time-series data of the S200 sample, facilitating model adaptation. Complete original management logs for the planting cycle of the plot to be evaluated are obtained through the regional agricultural big data platform and grower records. Key agricultural operations are extracted from the logs, organized by daily timestamps, and days with no operation are recorded as days without intervention, forming the original management time-series data.

[0062] For example, obtain the native management log: retrieve the native management log for plot T001 for the 2024-2025 period, which includes: sowing on September 29, 2024: 0.3 kg / mu; applying base fertilizer on October 16: 10 kg / mu of pure nitrogen + 8 kg / mu of phosphorus and potassium fertilizer; applying topdressing on November 23: 5 kg / mu of pure nitrogen; spraying imidacloprid on March 11, 2025: 30 ml / mu; extract the native management time-series data: organize it into time-series data by day, for example: September 29: no fertilization operation, no irrigation operation, sowing operation 0.3 kg / mu; October 16: fertilization operation: 10 kg / mu of pure nitrogen + 8 kg / mu of phosphorus and potassium fertilizer; no irrigation operation, no sowing operation; October 17-November 22: daily fertilization / irrigation / pesticide application operations are all recorded as none.

[0063] Secondly, real-time external source records of the plot to be evaluated are acquired and time-series data is generated to obtain real-time environmental time-series data and real-time crop growth time-series data. Real-time external source records refer to real-time monitoring data of the current planting cycle of the plot to be evaluated, obtained through IoT monitoring devices, UAV remote sensing, and meteorological platforms, including environmental monitoring data and crop growth monitoring data. Real-time environmental time-series data refers to daily data that has been time-series processed from real-time external environmental information, with dimensions consistent with the S200 sample environmental time-series data. Real-time crop growth time-series data refers to daily data that has been time-series processed from growth information obtained through UAV remote sensing and ground monitoring, including normalized vegetation index, leaf area index, etc., and its dimensions match those of the S200 sample growth time-series data. Real-time environmental data is collected through IoT sensors deployed on the plots and regional weather stations; crop growth data is collected through drone remote sensing field inspections every 10 days; the two types of data are organized according to daily timestamps, and non-daily data is completed into daily data through linear interpolation, forming real-time environmental time series data and real-time crop growth time series data.

[0064] For example, real-time environmental time-series data, partial data from April 1st to April 10th, 2025: April 1st: average daily temperature 18.5℃, cumulative sunshine 7.0 hours, rainfall 0mm, soil nitrogen content 22mg / kg, soil moisture 17%; April 5th: average daily temperature 20℃, cumulative sunshine 6.5 hours, rainfall 8mm, soil nitrogen content 21.5mg / kg, soil moisture 19%; real-time crop growth time-series data: April 1st (drone monitoring): normalized vegetation index 0.72, leaf area index 3.9; April 6th (interpolation completion): normalized vegetation index 0.73, leaf area index 3.95; April 10th (drone monitoring): normalized vegetation index 0.74, leaf area index 4.0.

[0065] Furthermore, the native management time-series data, the real-time environmental time-series data, and the real-time crop growth time-series data are time-series aligned. Time-series alignment refers to uniformly verifying the daily timestamps of the native management time-series data, the real-time environmental time-series data, and the real-time crop growth time-series data, using the sowing day of the current planting cycle as the starting point of the time axis, to ensure that the three types of data correspond one-to-one at the same time node, eliminating time misalignment caused by data collection delays and recording deviations. A continuous daily time axis is generated based on the sowing day of the current cycle; the three types of data are matched with this time axis respectively, and missing data nodes are marked; missing data is filled in using the same method as S200, environmental data is filled in using the average of adjacent plots in the same period, and growth data is filled in using interpolation; finally, a set of three types of time-series data with completely consistent timestamps is formed.

[0066] For example, the baseline timeline is from September 29, 2024 to April 10, 2025, with timestamps generated daily. The native management time-series data, real-time environmental time-series data, and real-time crop growth time-series data are mapped to each node on the timeline. Missing data is filled in: if environmental data is missing on December 5, 2024, it is filled in using environmental data from adjacent plots on that day; if crop growth data is missing on December 15, it is filled in using interpolated crop growth data from December 10 and December 20. Each timestamp corresponds to three complete types of data: environmental data, agricultural data, and crop growth data.

[0067] Then, the time-series alignment results are segmented according to the preset quality assessment interval to obtain multiple segmented time-series data with a temporal sequence relationship. The preset quality assessment interval refers to the 10-day assessment interval set in S200, which is consistent with the frequency of UAV field patrols and the sliding step size of model training to ensure that the assessment rhythm is adapted to the model. Segmented time-series data refers to continuous data segments formed by dividing the aligned complete time-series data into 10-day intervals. Each segment corresponds to an assessment cycle and has a clear temporal sequence relationship. Starting from the sowing day of the current planting cycle, the three types of data after time-series alignment are continuously segmented at 10-day intervals. Each segment contains 10 daily data points. If the last segment has less than 10 days, it is retained according to the actual number of days. The segmented data are numbered in chronological order to form multiple segmented time-series data.

[0068] For example, the data is segmented at 10-day evaluation intervals, starting 10 days after sowing on October 8, 2024, with each 10-day period forming a segment. The segmentation results are as follows: Segment P1: October 8-17, 2024, including environmental, agricultural, and crop growth data for this period; Segment P2: October 18-27, 2024... Segment P24: April 1-10, 2025. The segment numbers are sequentially increased to form an ordered segmented time-series dataset.

[0069] Furthermore, according to the aforementioned temporal sequence, multiple segmented time-series data are sequentially input into the mechanism-data coupling model for segmented quality assessment, obtaining a predicted quality fluctuation dataset. Segmented quality assessment refers to inputting each segmented time-series data into the mechanism-data coupling model one by one in chronological order. The mechanism-data coupling model outputs the predicted quality fluctuation value corresponding to that segment, i.e., the impact of the 10-day growth process on the final quality. The predicted quality fluctuation dataset is a collection formed by organizing the predicted quality fluctuation values ​​of all segments in chronological order, corresponding one-to-one with the segmented data. Each segmented time-series data is input into the mechanism-data coupling model in the order of segment number. The mechanism-data coupling model operates internally according to cascading logic: the upstream rapeseed mechanism model receives environmental and agricultural data and outputs implicit state variables; these are concatenated with the segmented growth data to form a fused feature vector, which is then input into the downstream quality mapping model; the quality mapping model outputs the predicted quality fluctuation value for that segment; and the predicted values ​​of all segments are organized in chronological order to form the predicted quality fluctuation dataset.

[0070] For example, the time-series data of segment P24 is input into the coupled model; the upstream mechanism model outputs the daily latent state variables of this segment: nitrogen stress factor 0.07-0.09, water stress factor 0.04-0.06, etc.; these are concatenated with the growth data of this segment: normalized vegetation index 0.72-0.74, and environmental data to form a 16-dimensional fused feature vector; the downstream quality mapping model outputs the predicted quality fluctuation value of this segment: oil content fluctuation +0.25%; thus forming a predicted quality fluctuation dataset. For example, some segment prediction results are: P22 fluctuation +0.18%, P23 fluctuation +0.22%, P24 fluctuation +0.25%; these are organized into a dataset by time series: {+0.05%, +0.08%, ..., +0.25%}.

[0071] Finally, the preset typical product quality data and the quality fluctuation dataset are cumulatively calculated to obtain the dynamic quality assessment result. The preset typical product quality data refers to the average product quality data of the plot to be assessed over the past five planting cycles, serving as the benchmark value for the cumulative quality fluctuation to ensure that the assessment result is consistent with the historical planting level of the plot. The cumulative calculation refers to using the typical product quality data as a benchmark, and accumulating the predicted quality fluctuation values ​​of each segment in chronological order to obtain the dynamic quality prediction value of each assessment node. The dynamic quality assessment result is a comprehensive assessment result that includes the quality prediction values ​​of each assessment node and the quality fluctuation trend, reflecting the change pattern of quality in the current cycle with the growth process. The process involves: determining the typical product quality data; accumulating the predicted quality fluctuation values ​​of each segment to the benchmark value in chronological order; calculating the dynamic quality prediction value at the end of each segment; and compiling the predicted values ​​and fluctuation trends of each node to form the dynamic quality assessment result.

[0072] For example, baseline data: typical oil content of plot T001 is 42.0%; cumulative calculation: predicted oil content after P22: 42.0% + cumulative fluctuation value from P1 to P22 3.18% = 45.18%; predicted oil content after P23: 45.18% + 0.22% = 45.40%; predicted oil content after P24: 45.40% + 0.25% = 45.65%. Dynamic quality assessment results: current predicted oil content is 45.65%, an increase of 3.65% compared to the typical value; fluctuation trend: the fluctuation range increases during the grouting period (March-April), and the quality continues to improve; simultaneously output predicted protein content of 21.6% and erucic acid content of 0.72%.

[0073] This includes obtaining the native management logs of the plot to be evaluated during the current planting cycle, and performing dynamic quality assessment based on the native management logs and the mechanism-data coupling model. The process also includes: The dynamic quality assessment result is compared with a preset first quality threshold. If the assessment result is lower than the first quality threshold, a quality warning signal is generated. The dynamic quality assessment result is compared with a preset second quality threshold. If the assessment result is higher than the second quality threshold, a cost warning signal is generated.

[0074] First, the dynamic quality assessment results are compared with a preset first quality threshold. If the assessment result is lower than the first quality threshold, a quality warning signal is generated. The first quality threshold refers to the minimum quality standard to ensure the quality of rapeseed products. It is set comprehensively based on industry standards, regional procurement requirements, and historical qualified quality data of the plots to be assessed. It covers core indicators such as oil content and erucic acid content and is the basis for judging whether the product has market value. The quality warning signal is a warning prompt generated when the dynamic quality assessment result is lower than the first quality threshold. It includes warning indicators, the gap between the current value and the threshold, and targeted field intervention suggestions to remind growers to adjust management measures in a timely manner to avoid the final quality failing to meet the standard. The preset first quality threshold is retrieved; the dynamic quality assessment results of each assessment node are compared with the first quality threshold one by one according to the indicator dimension. If the predicted value of any indicator is lower than the corresponding threshold, a quality warning is triggered; if multiple indicators fail to meet the standard, the warning priority is sorted from high to low according to the degree of failure. A standardized quality warning signal is generated, which clarifies the warning node, the non-compliant indicator and the gap, and the intervention suggestions, and is pushed to the grower's management terminal simultaneously.

[0075] For example, for the target variety Chuanyou 48, the first quality threshold is set as oil content ≥ 42.0%, erucic acid content ≤ 1.0%, and protein content ≥ 20.0%. Anomaly assessment scenario: On March 5, 2025, the dynamic quality assessment results for plot T001 were: oil content 41.8%, protein content 19.5%, and erucic acid content 0.8%, with both oil content and protein content below the first quality threshold. Warning signal generation: [Quality Warning - Segment P20] Current period quality warning for plot T001: Oil content 41.8%, below threshold 42.0%, difference 0.2%; Protein content 19.5%, below threshold 20.0%, difference 0.5%; Erucic acid content meets standard. Recommended measures: Apply 3 kg / mu of pure nitrogen within 3 days, and increase field irrigation frequency to once every 3 days to improve rapeseed grain filling efficiency.

[0076] Secondly, the dynamic quality assessment results are compared with a preset second quality threshold. If the assessment result is higher than the second quality threshold, a cost warning signal is generated. The second quality threshold is a critical quality value calculated based on the input-output ratio. That is, after the quality exceeds this value, the marginal cost of further quality improvement exceeds the marginal benefit. The indicator threshold is usually higher than the first quality threshold and is a key basis for balancing quality improvement and cost control. The cost warning signal is a warning notification generated when the dynamic quality assessment result is higher than the second quality threshold. It includes the threshold-exceeding indicator, the magnitude, and cost optimization suggestions, used to remind growers to reduce excessive input and avoid resource waste. The preset second quality threshold is retrieved, and the dynamic quality assessment results are compared with the second quality threshold one by one. If the predicted value of any indicator is higher than the corresponding threshold, a cost warning is triggered; if multiple indicators exceed the standard, the warning priority is sorted from high to low according to the cost increment. A standardized cost warning signal is generated, specifying the warning node, the exceeding indicator and its magnitude, and cost optimization suggestions, and is simultaneously pushed to the grower's management terminal.

[0077] For example, for the Sichuan Oil 48 variety, based on historical input data of the plot, a second quality threshold is set as oil content ≤ 45.0% and protein content ≤ 21.5%. Exceeding these values ​​requires additional fertilization / irrigation costs, and the market premium cannot cover the cost increase. Over-standard assessment scenario: As of April 10, 2025, the dynamic quality assessment results for plot T001 are: oil content 45.65%, protein content 21.6%, and erucic acid content 0.72%. Warning signal generation: [Cost Warning - Segment P24] Current cycle cost warning for plot T001 to be assessed: oil content 45.65%, higher than the threshold 45.0%, by 0.65%; protein content 21.6%, higher than the threshold 21.5%, by 0.1%. Calculations show that the additional fertilization cost corresponding to the current quality level is 80 yuan / mu, while the market premium for high-quality products is only 50 yuan / mu, resulting in an imbalance between input and output. Recommended measures: Stop applying topdressing fertilizer in the later stages, maintain the regular irrigation frequency, and avoid excessive improvement in quality that leads to cost waste.

[0078] In this embodiment of the invention, a dynamic, precise, and closed-loop management of rapeseed planting quality is achieved through a complete process of multi-source data integration, time-series alignment and segmentation, coupled model evaluation, and dual-threshold early warning. The original management time-series data of the current planting cycle of the plot to be evaluated is integrated with real-time environmental and growth time-series data. After time-series alignment and segmentation at preset intervals, the data is input into a mechanism-data coupling model to complete segmented quality fluctuation prediction. Then, combined with typical product quality data, a dynamic quality evaluation result is obtained through cumulative calculation. This breaks through the limitations of traditional post-harvest static evaluation due to its lag, enabling real-time tracking of the quality formation process. Simultaneously, by setting first and second dual quality thresholds, quality and cost early warning signals are triggered respectively. This not only promptly reminds growers to take field intervention measures to ensure product quality meets standards but also effectively avoids cost waste caused by excessive investment. Ultimately, a complete technical chain of data collection, dynamic evaluation, and early warning decision-making is formed, providing efficient and reliable technical support for precise management of the entire rapeseed planting cycle.

[0079] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects: This invention provides a method and system for assessing rapeseed planting quality based on big data analysis. First, multi-period, multi-dimensional data from source plots are screened based on intrinsic information similarity thresholds. This data, after standardization, forms a high-quality sample set, solving the problems of insufficient representativeness and data redundancy in traditional sample collection, and providing solid data support for model construction. Second, through time-series pruning, implicit state variable inversion, mechanistic model correction, and upstream-downstream model cascading, a mechanism-data coupled model is constructed, combining interpretability of physiological mechanisms with high data-driven accuracy. This overcomes the shortcomings of low prediction accuracy in pure mechanism models and weak generalization ability in pure data models, improving the scientific rigor and reliability of quality fluctuation prediction. Finally, the original management data and real-time monitoring data of the plots to be assessed are integrated, time-aligned, and segmented before being input into the coupled model to complete dynamic quality assessment. Combined with dual quality thresholds, quality and cost warning signals are generated, breaking the lag limitations of traditional post-harvest static assessment and achieving an upgrade from result detection to process control, while simultaneously achieving the dual goals of quality compliance and cost optimization. This has created a complete closed loop of data collection, model building, dynamic evaluation, and decision support, providing an efficient technological platform for large-scale and precision rapeseed cultivation.

[0080] Example 2, as Figure 2 As shown, this invention provides a rapeseed planting quality assessment system based on big data analysis, the system comprising: The sample data acquisition module 11 is used to collect sample data of the target rapeseed planting environment based on big data methods, wherein the sample data includes planting full record data and product quality data; The coupling model construction module 12 is used to construct a mechanism-data coupling model based on the sample data, wherein the mechanism-data coupling model includes a cascaded rapeseed mechanism model and a data-driven quality mapping model. The dynamic quality assessment module 13 is used to obtain the native management log of the plot to be assessed in the current planting cycle, and to perform dynamic quality assessment based on the native management log and the mechanism-data coupling model.

[0081] In one embodiment, the sample data acquisition module 11 is further configured to: Based on the intrinsic information of the land parcels to be evaluated, a set of parcels from the same source is determined; Based on the preset collection period constraints, the historical management logs of the land parcel to be evaluated and the set of parcels from the same source are extracted to obtain the sample data; The planting record data includes environmental data, agricultural management data, crop growth data, and corresponding product quality data from a preset historical planting cycle.

[0082] In one embodiment, the coupling model building module 12 is further configured to: Define the sample segmentation window based on the preset quality assessment interval; Based on the sample segmentation window, data is cropped starting from the latest time node of each sample data item to obtain cropped sample data. The cropped sample data is traversed, the full planting record data is extracted and data preprocessing is performed to form full time-series sample data; Traverse the cropped sample data, extract the first product quality data and the second product quality data corresponding to the beginning and end of the planting full record data, calculate the quality data difference, and obtain the sample quality fluctuation data; Using the sample quality fluctuation data as supervision and the sample full-time series data as input, the mechanism-data coupling model is constructed and trained.

[0083] The mechanism-data coupling model is constructed and trained using the sample quality fluctuation data as supervision and the sample full-time series data as input, including: Based on the sample crop growth time series data in the sample full-time series data, extract the training set of latent state variables; Based on the prior mechanism model, the rapeseed mechanism model is initialized and constructed. The rapeseed mechanism model is calibrated using the training set of the hidden state variables and the full-time data of the samples.

[0084] The implicit state variables include at least: nitrogen stress factor, water stress factor, heat stress factor, and grain allocation coefficient.

[0085] The mechanism-data coupling model, constructed and trained using the sample quality fluctuation data as supervision and the sample full-time series data as input, further includes: The full-time data of the samples and the training set of the latent state variables are time-series aligned and dimension-concatenated to obtain the sample fusion feature vector; A quality mapping model based on regression analysis is constructed, and the quality mapping model is iteratively trained using the sample fusion feature vector as training input and the sample quality fluctuation data as supervision. Using the quality mapping model as the downstream model and the rapeseed mechanism model as the upstream model, the models are cascaded to generate the mechanism-data coupling model.

[0086] The process of calibrating the rapeseed mechanism model using the latent state variable training set and the full-time data of the samples includes: Based on the full-time-series data of the sample, extract the time-series data of the sample environment and the time-series data of the sample agricultural management. The regulation response time of the target rapeseed variety is obtained by combining prior knowledge, and the time series data of the sample environment, the time series data of the sample agricultural management, and the training set of the latent state variables are time-series realigned according to the regulation response time. Based on the temporal realignment results, the sample environmental temporal data and the sample agricultural management temporal data, the rapeseed mechanism model is driven to perform verification simulation to obtain the implicit state variable verification set. The rapeseed mechanism model is calibrated by adjusting the model parameters with the goal of minimizing the residual between the latent state variable validation set and the latent state variable training set.

[0087] In one embodiment, the dynamic quality assessment module 13 is further configured to: Obtain the native management log of the plot to be evaluated in the current planting cycle, and extract the corresponding native management time series data; The real-time external source record information of the plot to be evaluated is obtained and time-series data is generated to obtain real-time environmental time-series data and real-time crop growth time-series data. Time-series alignment is performed on the native management time-series data, the real-time environment time-series data, and the real-time crop growth time-series data. The timing alignment results are segmented according to the preset quality assessment interval to obtain multiple segmented timing data with temporal sequence relationship; According to the aforementioned temporal sequence, multiple segmented time series data are sequentially input into the mechanism-data coupling model for segmented quality assessment to obtain a predicted quality fluctuation dataset; By combining preset typical product quality data with the quality fluctuation dataset, dynamic quality assessment results are obtained.

[0088] This includes obtaining the native management logs of the plot to be evaluated during the current planting cycle, and performing dynamic quality assessment based on the native management logs and the mechanism-data coupling model. The process also includes: The dynamic quality assessment result is compared with a preset first quality threshold. If the assessment result is lower than the first quality threshold, a quality warning signal is generated. The dynamic quality assessment result is compared with a preset second quality threshold. If the assessment result is higher than the second quality threshold, a cost warning signal is generated.

[0089] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0091] This specification and accompanying drawings are merely illustrative examples of the invention and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its scope. Therefore, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is intended to include these modifications and modifications.

Claims

1. A method for evaluating the quality of rapeseed cultivation based on big data analysis, characterized in that, include: Sample data of the target rapeseed planting environment were collected based on big data methods, wherein the sample data included planting full record data and product quality data; Based on the sample data, a mechanism-data coupling model is constructed, wherein the mechanism-data coupling model includes a cascaded rapeseed mechanism model and a data-driven quality mapping model; Obtain the native management logs of the plot to be evaluated in the current planting cycle, and perform dynamic quality assessment based on the native management logs and the mechanism-data coupling model.

2. The rapeseed planting quality assessment method based on big data analysis as described in claim 1, characterized in that, Sample data of the target rapeseed planting environment were collected using big data methods, including: Based on the intrinsic information of the land parcels to be evaluated, a set of parcels from the same source is determined; Based on the preset collection period constraints, the historical management logs of the land parcel to be evaluated and the set of parcels from the same source are extracted to obtain the sample data; The planting record data includes environmental data, agricultural management data, crop growth data, and corresponding product quality data from a preset historical planting cycle.

3. The rapeseed planting quality assessment method based on big data analysis as described in claim 1, characterized in that, Based on the sample data, a mechanism-data coupling model is constructed, wherein the mechanism-data coupling model includes a cascaded rapeseed mechanism model and a data-driven quality mapping model, comprising: Define the sample segmentation window based on the preset quality assessment interval; Based on the sample segmentation window, data is cropped starting from the latest time node of each sample data item to obtain cropped sample data. The cropped sample data is traversed, the full planting record data is extracted and data preprocessing is performed to form full time-series sample data; Traverse the cropped sample data, extract the first product quality data and the second product quality data corresponding to the beginning and end of the planting full record data, calculate the quality data difference, and obtain the sample quality fluctuation data; Using the sample quality fluctuation data as supervision and the sample full-time series data as input, the mechanism-data coupling model is constructed and trained.

4. The rapeseed planting quality assessment method based on big data analysis as described in claim 3, characterized in that, Using the sample quality fluctuation data as supervision and the sample full-time series data as input, the mechanism-data coupling model is constructed and trained, including: Based on the sample crop growth time series data in the sample full-time series data, extract the training set of latent state variables; Based on the prior mechanism model, the rapeseed mechanism model is initialized and constructed. The rapeseed mechanism model is calibrated using the training set of the hidden state variables and the full-time data of the samples.

5. The rapeseed planting quality assessment method based on big data analysis as described in claim 4, characterized in that, The rapeseed mechanism model is calibrated using the training set of the latent state variables and the full-time data of the samples, including: Based on the full-time-series data of the sample, extract the time-series data of the sample environment and the time-series data of the sample agricultural management. The regulation response time of the target rapeseed variety is obtained by combining prior knowledge, and the time series data of the sample environment, the time series data of the sample agricultural management, and the training set of the latent state variables are time-series realigned according to the regulation response time. Based on the temporal realignment results, the sample environmental temporal data and the sample agricultural management temporal data, the rapeseed mechanism model is driven to perform verification simulation to obtain the implicit state variable verification set. The rapeseed mechanism model is calibrated by adjusting the model parameters with the goal of minimizing the residual between the latent state variable validation set and the latent state variable training set.

6. The rapeseed planting quality assessment method based on big data analysis as described in claim 4, characterized in that, Using the sample quality fluctuation data as supervision and the sample full-time series data as input, the mechanism-data coupling model is constructed and trained, and further includes: The full-time data of the samples and the training set of the latent state variables are time-series aligned and dimension-concatenated to obtain the sample fusion feature vector; A quality mapping model based on regression analysis is constructed, and the quality mapping model is iteratively trained using the sample fusion feature vector as training input and the sample quality fluctuation data as supervision. Using the quality mapping model as the downstream model and the rapeseed mechanism model as the upstream model, the models are cascaded to generate the mechanism-data coupling model.

7. The rapeseed planting quality assessment method based on big data analysis as described in claim 1, characterized in that, The process of conducting dynamic quality assessment includes obtaining the native management logs of the plot to be assessed during the current planting cycle, and performing dynamic quality assessment based on the native management logs and the mechanism-data coupling model. This also includes: Obtain the native management log of the plot to be evaluated in the current planting cycle, and extract the corresponding native management time series data; The real-time external source record information of the plot to be evaluated is obtained and time-series data is generated to obtain real-time environmental time-series data and real-time crop growth time-series data. Time-series alignment is performed on the native management time-series data, the real-time environment time-series data, and the real-time crop growth time-series data. The timing alignment results are segmented according to the preset quality assessment interval to obtain multiple segmented timing data with temporal sequence relationship; According to the aforementioned temporal sequence, multiple segmented time series data are sequentially input into the mechanism-data coupling model for segmented quality assessment to obtain a predicted quality fluctuation dataset; By combining preset typical product quality data with the quality fluctuation dataset, dynamic quality assessment results are obtained.

8. The method for evaluating rapeseed planting quality based on big data analysis as described in claim 1, characterized in that, Obtain the native management logs of the plot to be evaluated during the current planting cycle, and perform dynamic quality assessment based on the native management logs and the mechanism-data coupling model. This process also includes: The dynamic quality assessment result is compared with a preset first quality threshold. If the assessment result is lower than the first quality threshold, a quality warning signal is generated. The dynamic quality assessment result is compared with a preset second quality threshold. If the assessment result is higher than the second quality threshold, a cost warning signal is generated.

9. The rapeseed planting quality assessment method based on big data analysis as described in claim 4, characterized in that, The implicit state variables include at least: nitrogen stress factor, water stress factor, heat stress factor, and grain allocation coefficient.

10. A rapeseed planting quality assessment system based on big data analysis, characterized in that, The system is used to implement the rapeseed planting quality assessment method based on big data analysis as described in any one of claims 1-9, the system comprising: The sample data acquisition module is used to collect sample data of the target rapeseed planting environment based on big data methods. The sample data includes planting full record data and product quality data. The coupling model construction module is used to construct a mechanism-data coupling model based on the sample data, wherein the mechanism-data coupling model includes a cascaded rapeseed mechanism model and a data-driven quality mapping model; The dynamic quality assessment module is used to obtain the native management logs of the plot to be assessed in the current planting cycle, and to perform dynamic quality assessment based on the native management logs and the mechanism-data coupling model.

Images