Method and system for evaluating wet tolerance of rape based on multi-source data fusion

Abstract

The application discloses a rapeseed moisture tolerance evaluation method and system based on multi-source data fusion, relates to the field of agricultural information technology, and synchronously acquires airspace remote sensing images, ground environment monitoring data and crop phenotype physiological indexes in the system operation; time and space alignment and feature extraction are carried out; a dynamic evaluation model based on deep learning is constructed, multi-dimensional features are fused to generate a whole growth period moisture tolerance evaluation index; and moisture tolerance grades are divided according to the index, and spatial visualization results and disaster loss suggestions are output. The system comprises data acquisition, preprocessing, core evaluation and result display modules, and integrates an expert decision support unit to realize closed-loop management. Through deep fusion of multi-source heterogeneous data and intelligent modeling, the application realizes real-time, quantitative and non-destructive accurate evaluation of rapeseed moisture tolerance, and significantly improves evaluation efficiency, accuracy and cross-region applicability.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, specifically to a method and system for evaluating the moisture tolerance of rapeseed based on multi-source data fusion. Background Technology

[0002] Due to the uneven spatial and temporal distribution of rainfall and climate fluctuations, waterlogging disasters are frequent in major rapeseed producing areas and have become a major abiotic stress factor limiting high and stable rapeseed yields. Accurate identification and evaluation of rapeseed's waterlogging tolerance are important prerequisites for the breeding of stress-resistant varieties, optimization of field irrigation and drainage, and scientific post-disaster assessment.

[0003] Among them, the rapeseed moisture tolerance evaluation technology based on multi-source data fusion represents the development direction of deep integration between smart agriculture and crop phenomics. This technology integrates diverse and heterogeneous information such as UAV remote sensing imagery, ground environmental monitoring data, vegetation morphology characteristics, and physiological and biochemical indicators to conduct comprehensive and non-destructive quantitative monitoring of rapeseed growth under moisture stress. This evaluation system, through the collaborative analysis of multi-dimensional data, aims to construct a comprehensive characterization model that reflects the evolution of rapeseed moisture tolerance throughout its entire growth cycle.

[0004] Current technologies for assessing rapeseed moisture tolerance still suffer from the following significant shortcomings: First, evaluation methods primarily rely on manual field observation and in vitro sampling, which is not only labor-intensive but also highly subjective, making rapid surveys difficult in large-scale planting areas. Second, data acquisition dimensions are too singular, often relying solely on surface information collected from a single sensor, lacking systematic correlation analysis between underground root development and aboveground canopy physiological changes. Third, data processing lacks a deep fusion mechanism, failing to effectively uncover the nonlinear coupling relationship between environmental meteorological data and crop physiological responses, resulting in insufficient accuracy and robustness of moisture tolerance assessment results. Finally, existing evaluation models have poor generalization ability, making it difficult to adapt to different soil environments, different developmental stages, and differentiated stress resistance performance among different varieties. These problems collectively lead to low efficiency and difficulty in quantifying rapeseed moisture tolerance assessment. Addressing the urgent technical challenges of limited assessment dimensions, low data fusion, strong reliance on manual labor, and poor model universality inherent in traditional methods, developing a rapeseed moisture tolerance assessment method and system based on multi-source data fusion has significant scientific research value and practical significance. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for evaluating the moisture tolerance of rapeseed based on multi-source data fusion, which solves the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for evaluating the moisture tolerance of rapeseed based on multi-source data fusion, comprising the following steps: S1. Multi-source heterogeneous raw data of rapeseed planting area are acquired synchronously through multi-source sensing terminals. The multi-source heterogeneous raw data includes airspace remote sensing image data, surface environment monitoring data and crop individual phenotypic physiological indicators. S2. Spatiotemporal alignment and feature processing are performed on multi-source heterogeneous raw data to extract multidimensional key phenotypic features reflecting the growth status of rapeseed and environmental risk factors characterizing the intensity of environmental adversity. S3. Construct a dynamic evaluation model for moisture tolerance based on deep learning, and nonlinearly couple multidimensional key phenotypic features with environmental risk factors to generate a moisture tolerance evaluation index for rapeseed throughout its entire growth period. S4. Based on the moisture resistance evaluation index, perform moisture resistance level classification and output evaluation results and disaster damage assessment suggestions containing spatial distribution information.

[0007] Preferably, step S2 specifically includes the following steps: S21, performing radiometric calibration and atmospheric correction on the spatial remote sensing image data, and extracting vegetation index features under different specific spectral bands; S22, performing temporal noise reduction and normalization processing on the surface environment monitoring data, and calculating the cumulative rainfall intensity and soil moisture gradient within a specific time period; S23, performing morphological fitting on the phenotypic physiological indicators of individual crops, and extracting morphological parameter features reflecting canopy structure and physiological activity; S24, using a spatial resampling algorithm to uniformly map the above features to a preset spatial geographic grid, achieving pixel-level spatial alignment of multimodal features.

[0008] Preferably, step S3 specifically includes the following steps: S31, establishing a multi-head attention mechanism model and calculating the contribution weights of different categories of features to the moisture tolerance evaluation; S32, using a temporal recurrent convolutional network to extract the dynamic response features of rapeseed to moisture stress at different growth stages; S33, using a fusion layer to deeply fuse the weighted static phenotypic features with the dynamic response features to construct a nonlinear mapping function; S34, using a preset moisture tolerance standard sample library as a benchmark, iteratively training and optimizing the parameters of the dynamic evaluation model.

[0009] Preferably, the spatial remote sensing image data includes multispectral image data and thermal infrared image data. The specific spectral bands cover the red light band, near-infrared band, and red edge band. Extraction of vegetation index features includes: calculating the normalized difference vegetation index based on the red light band and near-infrared band to characterize canopy biomass and coverage; and calculating the red edge position index based on the red edge band to characterize chlorophyll content and photosynthetic efficiency. Crop canopy temperature characteristics are extracted based on the thermal infrared image data, and the crop water stress index is calculated to reflect changes in transpiration intensity of rapeseed under wet conditions.

[0010] Preferably, the surface environment monitoring data is acquired through a sensor array deployed in the field. The calculation of environmental risk factors includes: collecting soil bulk density, soil moisture content, and groundwater level depth within a preset depth range; constructing a waterlogging accumulation index, the calculation logic of which is to integrate the difference in soil moisture content exceeding a preset saturation threshold over time within a preset time window. The environmental risk factors also incorporate the nonlinear relationship between surface runoff velocity and rainfall intensity to identify the distribution of low-lying areas susceptible to waterlogging.

[0011] Preferably, the individual crop phenotypic physiological indicators are collected using near-surface sensing devices, including plant height, stem diameter, leaf area index, and leaf water content. Morphological parameter extraction includes: using 3D point cloud reconstruction technology to obtain the spatial geometric topology of the rapeseed individual, and calculating canopy compactness and branching angle. Physiological indicator extraction also includes using a fluorescence detection device to obtain the maximum photon efficiency of the leaves, to assess the degree of damage to the type II reaction centers of the photosystem caused by wetland infestation.

[0012] Preferably, the dynamic evaluation model for moisture tolerance employs a multi-task learning framework. The first task branch is used to predict the current relative biomass loss rate of rapeseed; the second task branch is used to invert the degree of inhibition of respiration activity in the rapeseed root system. The dynamic evaluation model introduces a domain adaptation module, which improves the model's generalization evaluation ability across different geographical environments and different rapeseed varieties by minimizing the maximum uniform shear difference between the source domain sample distribution and the target domain sample distribution.

[0013] Preferably, the generation logic of the moisture tolerance evaluation index is based on a pre-defined fuzzy comprehensive evaluation matrix. This matrix assigns different weight coefficients to environmental risk factors and phenotypic characteristics according to different developmental stages of rapeseed. For example, during the seedling stage, the evaluation weights of soil moisture content and root morphology characteristics are increased; during the flowering stage, the weight ratios of canopy temperature and photosynthetic index are increased, thereby ensuring that the evaluation index can accurately reflect the differentiated sensitivity of rapeseed to moisture damage at different life stages.

[0014] Preferably, the moisture resistance level classification in step S4 adopts cluster analysis, dividing the moisture resistance evaluation index into four levels: extremely strong moisture resistance, moderate moisture resistance, sensitive, and extremely sensitive. The output evaluation results are displayed in the geographic information system in the form of a heat map. The color depth of the heat map is positively correlated with the moisture resistance score, realizing an intuitive presentation of the spatial heterogeneity of moisture resistance in large areas of farmland.

[0015] A rapeseed moisture tolerance evaluation system based on multi-source data fusion, used to implement the above-mentioned method, includes: a data acquisition module, which is equipped with an unmanned aerial vehicle (UAV) multispectral camera, a ground sensor node, and a portable physiological detection device, for real-time acquisition of multi-source heterogeneous raw data; a data preprocessing module, whose input is connected to the data acquisition module, and which integrates a coordinate transformation unit, a filtering and noise reduction algorithm, and a feature extraction operator, for generating aligned multi-dimensional feature vectors; a core evaluation module, which stores a pre-trained dynamic moisture tolerance evaluation model, configured to receive multi-dimensional feature vectors and calculate and generate a moisture tolerance evaluation index; and a result display module, which is connected to the core evaluation module, for converting the evaluation index into grade information and visualizing it using an electronic map.

[0016] Preferably, the UAV-borne multispectral camera in the data acquisition module is configured to perform automatic inspections according to a preset flight path overlap rate to ensure that the acquired image data has sufficient spatial resolution. The surface sensor nodes transmit data to the cloud server via a wireless communication protocol. The transmission process includes data encryption and verification steps to ensure the real-time nature and integrity of the environmental data.

[0017] Preferably, the core evaluation module also includes an expert decision support unit. This unit stores agronomic management plans for different moisture tolerance levels and can automatically generate drainage scheduling instructions, foliar fertilizer application suggestions, and replanting plans based on the current evaluation results, thereby achieving closed-loop management of evaluation and treatment.

[0018] This invention provides a method and system for evaluating the moisture tolerance of rapeseed based on multi-source data fusion, which has the following beneficial effects: (1) During system operation, this invention significantly improves the scientific rigor, accuracy, and operational efficiency of rapeseed moisture tolerance evaluation by constructing an evaluation system based on multi-source data fusion. First, the integration of multi-source heterogeneous data effectively compensates for the information gaps of single sensing methods. Through the deep interweaving of spatial, surface, and individual physiological indicators, a comprehensive representation dimension from microphysiology to macro-group is constructed, solving the structural problem of traditional methods' difficulty in linking underground root system and aboveground canopy changes. Second, this invention introduces deep learning and multi-head attention mechanisms, which can automatically mine the complex nonlinear coupling relationship between environmental meteorological data and crop physiological responses, greatly reducing the subjective uncertainty caused by human sensory bias. In summary, under the synergistic effect of multi-source data fusion and deep learning models, this invention achieves real-time, quantitative, and non-destructive monitoring of rapeseed moisture tolerance in complex field environments, providing unbiased data support for stress-resistant variety breeding and field management.

[0019] (2) By performing spatiotemporal alignment and feature processing, this invention ensures that information from different platforms and time dimensions can be fused in a unified spatial physical coordinate system. This pixel-level alignment mechanism avoids feature misalignment caused by heterogeneous data sources, making the correspondence between environmental stress intensity and crop phenotypic response more accurate and significantly improving the robustness of feature extraction. Using a temporal recurrent convolutional network to extract dynamic response features throughout the entire growth period can effectively identify the cumulative effect and recovery ability of rapeseed to waterlogging stress. Compared with single static observation, this dynamic modeling method can capture the evolution law of waterlogging tolerance, especially in different scenarios such as waterlogging during the seedling stage and continuous rainfall in the later stage, the model shows stronger discrimination accuracy.

[0020] (3) The introduction of a domain-adaptive module enables the moisture tolerance evaluation model to have cross-regional and cross-variety migration capabilities. By minimizing distribution differences, the model can automatically compensate for data fluctuations caused by soil type, climate background, and genetic differences, greatly enhancing the system's universality and generalization range, enabling its rapid promotion and application in large-scale planting areas. Based on pixel-level spatial heat map output, the evaluation results are deeply integrated with the geographic information system. This not only realizes the classification of moisture tolerance levels but also reveals the spatial distribution pattern of disaster situations in the field. Furthermore, through the closed-loop management suggestions provided by the expert decision support unit, managers can quickly locate high-risk areas and take differentiated irrigation and drainage measures, minimizing yield losses caused by waterlogging disasters.

[0021] (4) By using the red-edge position index and the crop water stress index, this invention can identify early stress signals in rapeseed earlier than traditional vegetation indices. During the latent period before visual symptoms appear, the model can issue early warnings through subtle changes in physiological activity and water status, providing a valuable time window for pre-disaster prevention and further enhancing the resilience of rapeseed production. The combination of three-dimensional point cloud reconstruction technology and fluorescence detection device in the system locks in the health status of rapeseed from both geometric and biophysical perspectives. This high-dimensional feature input, coupled with multi-task learning branches, makes the moisture tolerance evaluation no longer a simple classification process, but a comprehensive diagnostic process including loss analysis and vigor assessment, ensuring the completeness of the evaluation system. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall technical solution architecture of the rapeseed moisture tolerance evaluation method based on multi-source data fusion proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the dynamic evaluation model for moisture resistance in this invention; Figure 3This is a flowchart illustrating the logical flow of spatiotemporal alignment and feature processing of multi-source heterogeneous data in this invention. Figure 4 This is a schematic diagram of the overall architecture of the rapeseed moisture tolerance evaluation system based on multi-source data fusion proposed in this invention. Detailed Implementation

[0023] 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. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. Example 1

[0024] This invention provides a method for evaluating the moisture tolerance of rapeseed based on multi-source data fusion. Please refer to [link / reference]. Figure 1 This embodiment describes in detail a method for evaluating rapeseed's waterlogging tolerance based on multi-source data fusion, applied to the identification of rapeseed's stress resistance in high-standard farmland. In this embodiment, the evaluation system is deployed in a rapeseed variety screening base covering an area of ​​2,000 mu. This base has a relatively low-lying terrain, and rainfall distribution is extremely uneven throughout the rapeseed's growth period, making it highly susceptible to typical waterlogging stress.

[0025] Step S1 involves simultaneously acquiring multi-source heterogeneous raw data of the rapeseed planting area through multi-source sensing terminals. These terminals form a comprehensive sensing network covering the airspace, surface, and individual vegetation. At the airspace level, an industrial-grade UAV platform equipped with multispectral and thermal infrared sensors is used. This platform features a high-precision inertial measurement unit and positioning system, enabling high-precision flight at a preset altitude. The acquired airspace remote sensing image data includes red light, near-infrared, red-edge, and thermal infrared bands. The center wavelength of the red light band is set at 660 nm, the near-infrared band at 840 nm, and the red-edge band focuses on the 730 nm range where chlorophyll absorption changes dramatically. During image acquisition, the UAV strictly adheres to an 80% forward overlap and a 75% lateral overlap rate to ensure the geometric accuracy of the image stitching.

[0026] At the surface level, sensor arrays were deployed in a proportionally gridded layout within the experimental area. Each sensor node included multi-layered soil moisture sensors, soil bulk density sensors, and groundwater level monitors. The soil moisture sensors were buried at depths of 10 cm, 30 cm, and 50 cm below the surface to capture the dynamic moisture content of different soil layers in real time. Simultaneously, automatic tipping bucket rain gauges were installed in the field, with a sampling frequency set every 5 minutes to ensure accurate recording of sudden changes in rainfall intensity.

[0027] At the individual crop level, physiological indicators were collected from representative plants using portable near-ground sensing devices. This included acquiring plant height information using ultrasonic ranging sensors and obtaining point cloud data of the rapeseed canopy using a high-resolution laser 3D scanner. In addition, the photochemical efficiency of leaves was measured using a modulated chlorophyll fluorometer, and leaf senescence and abscission of rapeseed under wet stress were recorded using a portable leaf area meter.

[0028] Step S2 is executed to perform spatiotemporal alignment and feature processing on the multi-source heterogeneous raw data obtained above.

[0029] In step S21, radiometric calibration is performed on the spatial remote sensing image data. Specifically, a linear mapping relationship between image pixel values ​​and actual surface reflectance is established by placing a standard calibration plate with known reflectance on the ground, thereby eliminating interference caused by changes in light intensity. Subsequently, atmospheric correction is performed, using an aerosol correction model to eliminate the distortion of spectral information caused by atmospheric scattering and absorption effects. Based on this, vegetation index features under specific spectral bands are extracted. The normalized difference vegetation index is obtained by calculating the difference between the reflectance of the near-infrared band and the red band, and dividing it by the sum of their reflectance values. This index serves as a core indicator reflecting the canopy biomass, coverage, and vigorous growth of rapeseed. Further, using red-edge band information, the red-edge position index is calculated by identifying the wavelength position where the maximum spectral slope occurs. This index is highly sensitive to changes in chlorophyll content and can identify early signs of leaf chlorosis caused by waterlogging. In addition, based on the radiance values ​​obtained from the thermal infrared band and combined with synchronous ground temperature data, the crop water stress index is calculated. This index quantifies the temperature difference between the canopy and the air, and inversely reflects the degree of transpiration inhibition in rapeseed under water-saturated conditions.

[0030] In step S22, time-series noise reduction is performed on the surface environment monitoring data. A moving average filtering algorithm is used to eliminate abnormal jumps in values ​​caused by occasional sensor interference. Then, normalization is performed to map environmental parameters of different dimensions to a uniform range of 0 to 1. Key environmental risk factor extraction includes: real-time monitoring of soil moisture content within a preset 24-hour or 48-hour time window. When the soil moisture content exceeds a preset field capacity saturation threshold, the system automatically performs an integral operation over time on the difference exceeding this threshold, thereby generating a cumulative waterlogging index. This index can objectively quantify the physiological stress suffered by crop roots due to prolonged immersion in a hypoxic environment.

[0031] In step S23, morphological fitting is performed on the individual crop phenotypic physiological indicators. Using the acquired 3D point cloud data, background noise is removed through Euclidean clustering to reconstruct the spatial geometric topology of the rapeseed individuals. During this process, canopy compactness is calculated, which is the ratio of the actual volume occupied by the canopy to the volume of the smallest circumscribed sphere enclosing the canopy. Simultaneously, the angle between the primary branch and the main stem is precisely measured, i.e., the branching angle, to assess whether waterlogging has caused lodging due to excessive water absorption or decreased support. Regarding physiological activity, the maximum photon efficiency of the leaves is obtained through a fluorescence detection device, specifically by calculating the relative ratio between the original fluorescence and the maximum fluorescence of the type II reaction center of the photosystem to assess the substantial damage of waterlogging to the photosynthetic apparatus.

[0032] In step S24, a spatial resampling algorithm is executed. Since the spatial resolutions of UAV imagery, sensor node data, and individual phenotypic data differ, the system uniformly maps all features to a preset spatial geographic grid with sides of 0.5 meters. For sensor data acquired in discontinuous point arrays, a Kriging interpolation algorithm is used for spatial surface reconstruction to ensure spatial dimensional alignment of multimodal features at the pixel level.

[0033] Perform step S3 to construct a deep learning-based dynamic evaluation model for moisture resistance. This model employs an advanced multi-task learning architecture, aiming to simultaneously predict moisture resistance and analyze physiological gains and losses.

[0034] In step S31, a multi-head attention mechanism model is established. This model assigns different contribution weights to inputs of different dimensions by calculating the cross-correlation between feature vectors. For example, within 48 hours after a heavy rainfall event is identified, the attention mechanism automatically increases the weights of soil moisture content and canopy temperature features, while appropriately decreasing the weights of static morphological features. This dynamic weight allocation mechanism ensures that the model can focus on the indicators that are most explanatory to the current stress state.

[0035] In step S32, a temporal recurrent convolutional network is used to extract dynamic response features. This network structure is specifically designed to process temporally correlated sequential data. By introducing gated recurrent units into the hidden layers, the model can memorize the growth trend of rapeseed over the past 15 days and examine its current physiological performance within the entire developmental cycle. This allows the model to distinguish between fluctuations caused by instantaneous rainfall and irreversible damage caused by prolonged waterlogging, thereby capturing the dynamic response patterns of rapeseed to waterlogging stress.

[0036] In step S33, a nonlinear mapping function is constructed through a fusion layer. The fusion layer deeply interweaves the weighted static phenotypic features with the extracted dynamic response features. Using a multi-layer fully connected neural network, the model simulates the extremely complex biological feedback mechanism between environmental stress and crop response, ultimately calculating a value between 0 and 100, which is the moisture tolerance evaluation index for rapeseed throughout its entire growth period. The higher the index value, the stronger the crop's moisture tolerance.

[0037] In step S34, iterative training is performed based on a pre-defined standard sample library of wet tolerance. This sample library contains real performance records of 120 rapeseed varieties collected over the past 5 years under different levels of wet damage. By calculating the cross-entropy loss between the predicted evaluation index and the expert annotation results, the connection parameters within the model are continuously corrected using the backpropagation algorithm. Simultaneously, a domain adaptation module is introduced. This module minimizes the difference between the data distribution of the current evaluation area and the distribution of the standard sample library by executing distribution alignment logic, thereby eliminating systematic biases caused by differences in soil type or climate year, ensuring that the model has extremely strong cross-scenario generalization ability.

[0038] Step S4 is executed to classify the moisture resistance level based on the moisture resistance evaluation index. Using a hard classification logic in cluster analysis, the evaluation index is divided into four levels: an index between 85 and 100 is defined as extremely strong moisture resistance; between 65 and 84 is defined as moderate moisture resistance; between 40 and 64 is defined as sensitive; and below 40 is defined as extremely sensitive.

[0039] Finally, the system outputs evaluation results containing spatial distribution information. These results are embedded in the farmland geographic information system as a high-resolution heat map. The heat map uses a gradient from dark blue to dark red; dark blue areas represent stable-yield areas with excellent moisture tolerance, while dark red areas visually indicate severely affected areas due to poor drainage or water accumulation. Simultaneously, the system, in conjunction with an expert decision support unit, outputs disaster damage assessment recommendations. For areas classified as extremely sensitive or sensitive, the system automatically retrieves weather forecast data. If rainfall is predicted within the next three days, a secondary drainage warning is immediately sent to the management terminal, and foliar fertilizer application recommendations with specific nitrogen and potassium fertilizer ratios are generated to enhance crop stress recovery. Example 2

[0040] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, this embodiment focuses on describing a moisture tolerance evaluation system and method applied in large-scale commercial rapeseed production areas. Compared with Embodiment 1, this embodiment places greater emphasis on large-scale efficiency in data acquisition and focuses on rapid prediction of population yield loss in model logic.

[0041] In step S1, for wide-area planting areas, the main reliance is on medium-to-high resolution satellite remote sensing data and regional automatic weather station data. Although its spatial resolution is lower than that of drones, its coverage area can reach tens of thousands of hectares. The airspace image data uses multi-temporal spectral fusion data, which can monitor the entire process of rapeseed from the seedling stage, budding stage to flowering stage.

[0042] In step S2, feature extraction focuses on macroscopic environmental risk indicators. In addition to rainfall, this embodiment introduces surface runoff velocity characteristics. By analyzing digital elevation model data, the catchment area and slope coefficient of each grid cell are calculated, thereby identifying potentially flood-prone areas. This topographic-based preprocessing method allows the system to spatially model potential risks before rainfall occurs.

[0043] In the evaluation logic of step S3, the first task branch is set to predict the current relative biomass loss rate of rapeseed. The model calculates the proportion of missing biomass by comparing the spectral curve of the affected area with the baseline curve of the healthy growing area in the same period of history using a nonlinear regression algorithm. The second task branch focuses on inverting the degree of inhibition of respiration activity of rapeseed roots. Since the most direct damage of waterlogging is the obstruction of aerobic respiration caused by root hypoxia, the model indirectly assesses the survival status of the root system by establishing a correlation between the water transpiration rate of the aboveground parts and the oxygen diffusion coefficient of the underground parts. This two-branch structure allows the evaluation index to include not only the current growth assessment but also the trend prediction of future yield.

[0044] In the output stage of step S4, the system achieves closed-loop management of evaluation and remediation. The results display module not only generates heat maps but also automatically calculates the required drainage dynamic parameters for the area based on the severity of damage in each grid, such as the required pump power and estimated drainage duration. This engineered recommendation directly connects to the intelligent irrigation and drainage facilities in the field, realizing an automated response from monitoring to execution. Example 3

[0045] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, this embodiment describes a refined evaluation scenario for rapeseed stress resistance breeding research. In this scenario, the evaluation accuracy is required to reach the fine structure level of individual vegetation, with the aim of screening out superior materials carrying strong moisture-resistant genes from thousands of genetic populations.

[0046] In the data acquisition phase of step S1, the core sensing terminal not only includes conventional equipment but also incorporates a highly sensitive chlorophyll fluorescence dynamics imaging system and a root scanning device. These devices can capture early physiological stress signals that are invisible to the naked eye. For example, fluorescence imaging can show the non-uniform distribution of photosynthetic activity on the leaf surface, thus detecting signs of waterlogging damage 2 to 3 days before the leaves begin to yellow.

[0047] In the environmental characteristic calculation of step S22, this embodiment uses a smaller time step for environmental risk analysis. The system calculates not only cumulative rainfall but also the gradient rate of change of soil moisture content. By monitoring the rate of soil moisture drainage after heavy rainfall, the interference of the drainage performance of the experimental plot on the experimental results is assessed and removed as a correction factor in subsequent evaluation.

[0048] In the dynamic evaluation model of step S3, a fuzzy comprehensive evaluation matrix based on developmental stages is introduced. During the rapeseed seedling stage, the weight coefficients of soil bulk density and root morphology parameters are increased to 0.6, because root development is most sensitive to water at this stage. After entering the flowering stage, the weights of canopy temperature and red edge index are increased to focus on monitoring the impact of water stress on reproductive growth. This parameter adjustment based on biological developmental rhythms gives the evaluation index extremely high physiological explanatory power.

[0049] In the application of the results in step S4, the system not only outputs the moisture tolerance level, but also generates a genetic gain assessment report for the breeding purpose. This report can compare the performance differences of different breeding materials under the same moisture damage intensity, and automatically record the materials with outstanding performance into the stress resistance gene library, providing accurate phenotypic data support for subsequent molecular marker development and breeding decisions.

[0050] The present invention proposes a rapeseed moisture tolerance evaluation system based on multi-source data fusion, the internal structure of which is closely integrated with the above-mentioned method and process.

[0051] The data acquisition module is equipped with a high-performance embedded processor to coordinate the synchronous operation of the drone, surface sensors, and physiological monitoring equipment. This module has a built-in timing protocol that ensures the timestamp error of all sensor data is controlled within 10 milliseconds, thereby guaranteeing strict alignment of multi-source data in the time dimension.

[0052] The data preprocessing module integrates a coordinate transformation unit. This unit can uniformly project raw data acquired from different sensors to the Gauss-Kruger coordinate system or the global latitude and longitude coordinate system. The internal filtering and noise reduction algorithm uses multidimensional Gaussian smoothing logic, which can effectively remove background noise caused by illumination fluctuations or poor sensor contact. The feature extraction operator is a set of highly optimized algorithms that can quickly calculate the aforementioned vegetation indices and environmental factors from massive amounts of raw imagery and time-series data.

[0053] The core evaluation module serves as the command center of the entire system, storing a dynamic moisture tolerance evaluation model trained through tens of thousands of iterations. This module is equipped with a dedicated computing unit capable of supporting rapid inference by deep learning networks. Its internal task allocation logic ensures the parallel execution of multi-head attention computation and recurrent convolution processing, significantly improving the efficiency of evaluation index generation. Furthermore, the expert decision support unit embedded within the core evaluation module is not only a rule base but also an inference engine based on historical cases, capable of generating targeted agronomic suggestions based on real-time moisture tolerance indices.

[0054] The results display module utilizes Geographic Information System (GIS) components to transform complex mathematical evaluation results into intuitive spatial geometric graphics. It supports the overlay display of multi-layered data, allowing users to click on any pixel on the heatmap to view the corresponding original feature parameters and historical evaluation curves, thus achieving traceability and visualization of the evaluation process. Example 4

[0055] This embodiment is an explanation based on Embodiment 3. Please refer to it. Figure 1 Specifically, in the implementation of this invention, all logical judgments strictly adhere to physical and biological principles. For example, when calculating vegetation indices, the system pre-sets a physical threshold range for reflectance; any reflectance data exceeding the 0-1 range is considered invalid, triggering an abnormal data compensation mechanism. When processing soil moisture content, the system combines soil texture parameters to correct bulk density, ensuring that the extracted environmental risk factors accurately reflect the intensity of oxygen deficiency pressure experienced by the root system.

[0056] In summary, this invention, through a three-dimensional perception framework integrating air, ground, and individual elements, combined with a multi-head attention mechanism and a multi-task learning deep learning model, fundamentally changes the traditional evaluation of rapeseed waterlogging tolerance, which relies on manual labor, has a single dimension, and is highly subjective. It not only achieves real-time, quantitative, and non-destructive monitoring of rapeseed growth under waterlogging stress, but also constructs a scientific, rigorous, and highly universal evaluation system through pixel-level spatial alignment and non-linear coupling of multi-dimensional features. The establishment of this system provides a key technical path for the precise breeding of stress-resistant rapeseed varieties and also provides a decision-making basis for modern agricultural disaster prevention and mitigation, and precision irrigation and drainage, demonstrating significant economic, social, and ecological benefits.

[0057] In the above embodiments of the present invention, through the collaborative processing of multi-source heterogeneous data, the system can automatically identify differences in characteristic responses under different climatic backgrounds. For example, in a low-temperature and rainy environment, the model will automatically increase its sensitivity to chlorophyll fluorescence indicators to cope with possible low-temperature stress; while in a high-temperature and high-humidity environment, it is more inclined to identify the potential risk of disease outbreaks through thermal infrared data. This adaptive logical adjustment ensures that the evaluation method can maintain extremely high evaluation accuracy in different rapeseed producing areas across the country.

[0058] Regarding system operational security, the surface sensor nodes employ low-power wide-area network wireless communication technology and implement stringent data verification and retransmission mechanisms. When data packet corruption or loss is detected, the cloud server immediately sends a retransmission command to ensure the integrity of the time-series data chain required for evaluation. Simultaneously, the core evaluation module possesses self-diagnostic capabilities; when input features are severely missing or sensors fail, the system automatically degrades to a single data source evaluation mode and issues an accuracy degradation alarm to the user.

[0059] Furthermore, through the closed-loop management provided by the expert decision support unit, this system achieves a seamless connection from "problem identification" to "problem resolution." The drainage scheduling instructions generated by the system can estimate the flow rate based on the actual carrying capacity of farmland drainage ditches, avoiding secondary soil erosion or resource waste caused by blind drainage. The foliar fertilizer application suggestions provide precise element ratio schemes based on the current physiological deficit state of rapeseed, maximizing fertilizer utilization efficiency.

[0060] In practical applications, this invention is not only suitable for variety selection in research institutions, but also greatly empowers the production management of large-scale farm owners. Through a mobile application, managers can view the moisture tolerance heat map of farmland anytime, anywhere, and receive automatically generated early warning alerts. This real-time interactive method significantly shortens the response time after a disaster, minimizing the impact of waterlogging disasters on yields. Example 5

[0061] For a method to evaluate rapeseed's moisture tolerance based on multi-source data fusion, please refer to [reference needed]. Figures 1 to 4 Specifically, it includes the following steps: S1. Acquire multi-source heterogeneous raw data of rapeseed planting area synchronously through multi-source sensing terminals; wherein, the multi-source heterogeneous raw data includes airspace remote sensing image data, surface environment monitoring data and crop individual phenotypic physiological indicators; S2. The multi-source heterogeneous raw data is subjected to spatiotemporal alignment and feature processing to extract multidimensional key phenotypic features reflecting the growth status of rapeseed and environmental risk factors characterizing the intensity of environmental adversity. Specifically, radiometric calibration and atmospheric correction are performed on the spatial remote sensing image data to extract vegetation index features under different specific spectral bands; temporal denoising and normalization processing is performed on the surface environmental monitoring data to calculate the cumulative rainfall intensity and soil moisture gradient within a specific time period; morphological fitting is performed on the individual crop phenotypic physiological indicators to extract morphological parameter features reflecting canopy structure and physiological activity; and a spatial resampling algorithm is used to uniformly map the above features to a preset spatial geographic grid to achieve pixel-level spatial alignment of multimodal features. S3. Construct a dynamic evaluation model for moisture tolerance based on deep learning, and nonlinearly couple the multidimensional key phenotypic features with the environmental risk factors to generate a moisture tolerance evaluation index for rapeseed throughout its entire growth period; specifically: establish a multi-head attention mechanism model to calculate the contribution weights of different categories of features to the moisture tolerance evaluation; use a temporal recurrent convolutional network to extract the dynamic response features of rapeseed to moisture stress at different growth stages; deeply fuse the weighted static phenotypic features with the dynamic response features through a fusion layer to construct a nonlinear mapping function; and iteratively train and optimize the dynamic evaluation model for moisture tolerance using a pre-set moisture tolerance standard sample library as a benchmark. S4. Based on the moisture resistance evaluation index, perform moisture resistance level classification and output evaluation results and disaster damage assessment suggestions containing spatial distribution information.

[0062] In this embodiment, the present invention constructs a multi-source data-driven dynamic evaluation method for rapeseed moisture tolerance, achieving accurate identification and quantitative assessment of rapeseed growth status under moisture stress. First, by collaboratively acquiring and fusing spatial remote sensing image data, surface environmental monitoring data, and individual crop phenotypic physiological indicators, multidimensional information about the rapeseed growth process can be comprehensively depicted from the macro-environment to the micro-plant level, improving the completeness and reliability of data sources. Second, through spatiotemporal alignment and multimodal feature extraction techniques, data from different sources and scales are uniformly mapped to the same spatial geographic grid, achieving pixel-level fusion of multi-source information, thereby significantly improving the accuracy of identifying rapeseed growth status and environmental stress intensity. Furthermore, by introducing a multi-head attention mechanism and a temporal recurrent convolutional network to construct a dynamic evaluation model for moisture tolerance, the dynamic response characteristics of rapeseed to moisture stress throughout its entire growth period can be effectively captured, realizing nonlinear coupling analysis between environmental factors and phenotypic characteristics, and improving the intelligence and dynamism of moisture tolerance evaluation. Ultimately, by using the moisture tolerance evaluation index and its spatial distribution results, the moisture tolerance level of rapeseed can be finely classified, and a scientific basis for decision-making can be provided for moisture damage risk warning, disaster damage assessment, variety selection and field management, thereby improving the precision of rapeseed production management and disaster resistance.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the moisture tolerance of rapeseed based on multi-source data fusion, characterized in that: Includes the following steps: S1. Simultaneously acquire multi-source heterogeneous raw data of rapeseed planting areas through multi-source sensing terminals; among which, multi-source heterogeneous raw data includes airspace remote sensing image data, surface environment monitoring data, and individual crop phenotypic and physiological indicators. S2. Spatiotemporal alignment and feature processing are performed on multi-source heterogeneous raw data to extract multidimensional key phenotypic features reflecting the growth status of rapeseed and environmental risk factors characterizing the intensity of environmental adversity. Specifically, radiometric calibration and atmospheric correction are performed on the spatial remote sensing image data to extract vegetation index features under different specific spectral bands; temporal denoising and normalization processing is performed on the surface environmental monitoring data to calculate the cumulative rainfall intensity and soil moisture gradient within a specific time period; morphological fitting is performed on the phenotypic physiological indicators of individual crops to extract morphological parameter features reflecting canopy structure and physiological activity; and a spatial resampling algorithm is used to uniformly map the above features to a preset spatial geographic grid to achieve pixel-level spatial alignment of multimodal features. S3. Construct a deep learning-based dynamic evaluation model for moisture tolerance, nonlinearly coupling multidimensional key phenotypic features with environmental risk factors to generate a moisture tolerance evaluation index for rapeseed throughout its entire growth period. Specifically: establish a multi-head attention mechanism model to calculate the contribution weights of different categories of features to the moisture tolerance evaluation; use a temporal recurrent convolutional network to extract the dynamic response features of rapeseed to moisture stress at different growth stages; deeply fuse the weighted static phenotypic features with the dynamic response features through a fusion layer to construct a nonlinear mapping function; and iteratively train and optimize the dynamic evaluation model for moisture tolerance using a pre-defined moisture tolerance standard sample library as a benchmark. S4. Based on the moisture resistance evaluation index, perform moisture resistance level classification and output evaluation results and disaster damage assessment suggestions containing spatial distribution information.

2. The method for evaluating rapeseed moisture tolerance based on multi-source data fusion according to claim 1, characterized in that: The process of extracting vegetation index features in step S21 is as follows: Based on the red band reflectance and near-infrared band reflectance in the spatial remote sensing image data, the normalized difference vegetation index is calculated to characterize the canopy biomass and coverage of rapeseed; the red edge position index is calculated based on the red edge band, and by identifying the wavelength position where the maximum spectral slope is located, the decay state of chlorophyll content and photosynthetic efficiency is characterized. The radiance value is obtained based on thermal infrared image data, and the crop water stress index is calculated by combining the air temperature data collected simultaneously on the ground. By quantifying the temperature difference between the canopy temperature and the air temperature, the degree of inhibition of transpiration intensity of rapeseed under water saturation environment is inverted.

3. The method for evaluating rapeseed moisture tolerance based on multi-source data fusion according to claim 1, characterized in that: The process of calculating environmental risk factors in step S22 is as follows: The system collects soil bulk density, soil moisture content, and groundwater level depth within a preset depth range using a sensor array deployed in the field. The preset depth range includes the depth of multiple discrete soil layers below the surface. The cumulative waterlogging index is constructed by monitoring soil moisture content in real time within a preset time window. When the soil moisture content exceeds the preset field water holding capacity saturation threshold, the difference exceeding the saturation threshold is integrally calculated over time. The environmental risk factor also incorporates the nonlinear correlation characteristics between surface runoff velocity and rainfall intensity extracted based on the digital elevation model to identify the distribution of low-lying catchment areas in farmland that are susceptible to waterlogging.

4. The method for evaluating rapeseed moisture tolerance based on multi-source data fusion according to claim 1, characterized in that: The process of extracting morphological parameter features in step S23 is as follows: Portable near-ground sensing devices were used to collect data on plant height, stem diameter, leaf area index, and leaf water content. The spatial geometric topology of individual rapeseed plants is obtained by using 3D point cloud reconstruction technology. Background noise is removed and the rapeseed canopy is reconstructed by Euclidean clustering algorithm. The canopy compactness is calculated. The canopy compactness is defined as the ratio of the actual spatial volume occupied by the canopy to the volume of the smallest circumscribed sphere that encloses the canopy. The branching angle between the primary branch and the main stem is measured to assess the risk of crop lodging caused by waterlogging. The maximum photon efficiency of the leaves is obtained using a fluorescence detection device, and the degree of damage to the photosynthetic apparatus caused by waterlogging is assessed by calculating the relative ratio between the original fluorescence and the maximum fluorescence of the photosystem reaction center.

5. The method for evaluating rapeseed moisture tolerance based on multi-source data fusion according to claim 1, characterized in that: The process of pixel-level spatial alignment of multimodal features in step S24 is as follows: For image data, sensor data and individual phenotypic data with different spatial resolutions, the Kriging interpolation algorithm is used to reconstruct the spatial surface of discontinuous point-collected surface environmental monitoring data. All extracted features are resampled into geographic grid cells of uniform size to ensure that data features from different sources are overlaid and fused in a unified spatial physical coordinate system.

6. The method for evaluating rapeseed moisture tolerance based on multi-source data fusion according to claim 1, characterized in that: The dynamic evaluation model for moisture resistance employs a multi-task learning framework, which includes: The first task branch is configured to predict the current relative biomass loss rate of rapeseed by comparing the spectral curve of the disaster area with the baseline curve of the healthy growing area in the same period in history. The second task branch is configured to establish the correlation logic between the water transpiration rate of the aboveground parts and the oxygen diffusion coefficient of the underground parts, and to invert the degree of inhibition of respiratory activity of rapeseed roots. The dynamic evaluation model for moisture tolerance also introduces a domain adaptive module, which minimizes the maximum uniform shear difference between the source domain sample distribution and the target domain sample distribution by executing distribution alignment logic, thereby improving the generalization evaluation capability of the dynamic evaluation model for moisture tolerance in different geographical environments and different rapeseed varieties.

7. The method for evaluating rapeseed moisture tolerance based on multi-source data fusion according to claim 1, characterized in that: The generation logic of the moisture resistance evaluation index is based on a set of preset fuzzy comprehensive evaluation matrices; The fuzzy comprehensive evaluation matrix assigns different weight coefficients to environmental risk factors and multidimensional key phenotypic features according to different developmental stages of rapeseed. During the rapeseed seedling stage, the evaluation weight coefficients of soil moisture content, soil bulk density, and root morphology characteristics were increased; during the rapeseed flowering stage, the weight coefficients of canopy temperature, red edge position index, and photosynthetic index were increased. By dynamically adjusting the weighting coefficients, the moisture tolerance evaluation index reflects the differentiated sensitivity of rapeseed to moisture damage at different life stages.

8. The method for evaluating rapeseed moisture tolerance based on multi-source data fusion according to claim 1, characterized in that: The process of performing moisture resistance rating classification and result output in step S4 is as follows: Using the hard classification logic in cluster analysis, the calculated moisture resistance evaluation index is divided into four levels: extremely strong moisture resistance, moderate moisture resistance, sensitive, and extremely sensitive. The moisture tolerance level is displayed in the geographic information system in the form of a high-resolution heat map. The color depth of the heat map is positively correlated with the value of the moisture tolerance evaluation index, which is used to present the spatial heterogeneity of moisture tolerance in farmland. The disaster damage assessment recommendations combine real-time weather forecast data and automatically generate agronomic management plans that include drainage scheduling instructions and foliar fertilizer application ratio schemes for areas classified as sensitive or extremely sensitive.

9. The rapeseed moisture tolerance evaluation system based on multi-source data fusion according to claim 1, applied to the rapeseed moisture tolerance evaluation method based on multi-source data fusion according to any one of claims 1 to 8, characterized in that: The system includes: The data acquisition module is equipped with an unmanned aerial vehicle-borne multispectral camera, a ground sensor node, and a portable physiological detection device. It is used to coordinate the synchronous operation of each device through an embedded processor to acquire multi-source heterogeneous raw data in real time. The data preprocessing module, whose input end is connected to the data acquisition module, integrates a coordinate transformation unit, a filtering and noise reduction algorithm, and a feature extraction operator. It is used to uniformly project the acquired raw data to a preset coordinate system and perform multidimensional Gaussian smoothing noise reduction to generate aligned multidimensional feature vectors. The core evaluation module stores a pre-trained dynamic evaluation model for moisture resistance, is equipped with computing power units to support parallel inference of multi-head attention mechanism and recurrent convolutional network, receives multi-dimensional feature vectors and calculates and generates moisture resistance evaluation index. The results display module is connected to the core evaluation module. It uses geographic information system components to convert the moisture resistance evaluation index into grade information and combines it with an electronic map to output a heat map visualization.

10. The rapeseed moisture tolerance evaluation system based on multi-source data fusion according to claim 1, characterized in that: The core evaluation module also includes an expert decision support unit, which stores a base of agronomic management rules for different moisture tolerance levels. The expert decision support unit is configured to automatically retrieve and match the corresponding irrigation and drainage dynamic parameters and stress recovery fertilizer formula based on the current moisture tolerance evaluation index and spatial distribution information, so as to realize closed-loop management of evaluation and treatment. The system also has a self-diagnostic function. When it detects that the input feature vector is missing or the sensor is malfunctioning, it automatically degrades to a single data source evaluation mode and outputs an accuracy decay alarm.

Images