A field crop health dynamic evaluation system based on multi-modal agricultural data fusion

Abstract

The application discloses a kind of field crop health dynamic evaluation systems based on multi-modal agricultural data fusion, it is related to agricultural information technology field.The system includes multi-modal data acquisition module, edge preprocessing module, cloud fusion analysis platform and decision push module.Through remote sensing unit, near-ground sensing unit and soil detection unit, multi-source heterogeneous data are synchronously acquired, and time-space alignment and feature extraction are realized by using edge preprocessing module;Multi-dimensional agricultural big data set is constructed in cloud platform, and crop growth, pest risk and yield trend are analyzed by combining machine learning model to generate crop health comprehensive index;The decision push module integrates market data, outputs fertilization, irrigation and plant protection scheme considering ecological and economic benefits, and supports variable operation;It breaks the information barrier of single data source, solves the problem of one-sided diagnosis and high misjudgment rate in traditional evaluation, and realizes the digital management of agricultural production from precise perception to intelligent decision-making to closed-loop control.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, specifically providing a field crop health dynamic assessment system based on multimodal agricultural data fusion. Background Technology

[0002] In the development of modern precision agriculture, real-time and accurate assessment of crop growth in the field is a prerequisite for scientific water and fertilizer management. Currently, agricultural producers mainly rely on data from a single source for agricultural decision-making, such as monitoring soil temperature and humidity solely through ground-based IoT sensors or obtaining vegetation cover indices only through satellite remote sensing. However, soil sensors can only reflect local physical indicators underground and cannot perceive the physiological stress responses of crops above ground; while remote sensing images, although covering a wide area, are easily obscured by clouds and lack information related to the underground root environment, resulting in biased and delayed assessments of crop health and making it difficult to cope with complex and ever-changing field environments.

[0003] Furthermore, most existing crop assessment systems lack the ability to deeply integrate heterogeneous data. Data from different modalities (such as spectral data, environmental parameters, and soil physicochemical properties) are typically stored and processed independently, resulting in poor temporal synchronization and spatial resolution mismatch. This makes it difficult for the system to accurately distinguish whether crop stress is caused by water shortage or pests and diseases, easily leading to misjudgments. For example, nitrogen deficiency and root diseases in crops may exhibit similar spectral characteristics. If an accurate diagnosis cannot be made based on a single data source, it can lead to errors in irrigation or fertilization decisions, resulting not only in resource waste but also potentially affecting crop yield and quality.

[0004] Therefore, there is an urgent need for a field crop health dynamic assessment system that can integrate multimodal agricultural data and achieve spatiotemporal alignment and feature-level fusion. This system should break down information barriers from single data sources, dynamically and accurately quantify crop health status through multi-dimensional cross-validation, and adaptively generate differentiated field management strategies based on the assessment results, thereby addressing the shortcomings of existing technologies such as single assessment dimensions and low diagnostic accuracy. Summary of the Invention

[0005] To address the aforementioned problems, the present invention proposes the following technical solution: a field crop health dynamic assessment system based on multimodal agricultural data fusion, comprising: The multimodal data acquisition module is used to simultaneously acquire remote sensing image data, near-ground environmental data, multidimensional soil data, and historical production data of the target area; An edge preprocessing module, connected to the multimodal data acquisition module, is used to perform spatiotemporal alignment, cleaning, and feature extraction on the multimodal data; The cloud-based fusion analysis platform communicates with the edge preprocessing module to construct a multi-source agricultural big data set and uses machine learning algorithms to intelligently analyze crop growth status and pest and disease risks, generating a dynamic crop health assessment report. The decision push module is used to visualize the generated accurate decision-making solutions and push them to user terminals or agricultural machinery equipment.

[0006] Preferably, the multimodal data acquisition module includes: The remote sensing unit is configured as a drone or satellite remote sensing device equipped with a multispectral camera to acquire information on the spatial distribution of farmland and spectral data on crop growth. The near-ground sensing unit is configured as an IoT base station deployed in the field to collect data on air temperature and humidity, light intensity, rainfall, and CO2 concentration. The soil detection unit is configured as a composite probe embedded in the root layer to detect soil volumetric water content, nutrient content, pH value and electrical conductivity; Data storage unit, used to store historical agricultural production records and market data.

[0007] Preferably, the edge preprocessing module includes: The spatiotemporal alignment submodule is used to assign a unified time label to heterogeneous data from different sources and map them to the same geographic coordinate system; The data cleaning submodule is used to remove outliers and missing values, and to correct sensor drift data. The feature extraction submodule is used to extract vegetation index features, environmental stress features, and soil fertility features from the raw data.

[0008] Preferably, the cloud-based fusion analysis platform includes a data fusion layer and a model inference layer; The data fusion layer is used to perform multi-source data fusion modeling on the extracted features and construct a multi-source agricultural big data set that includes spatiotemporal dimensions, environmental dimensions, and growth dimensions. The model inference layer has multiple built-in machine learning models for different crop growth cycles, which are used to analyze the current growth status of crops, predict the risk of pest and disease outbreaks, and yield trends.

[0009] Preferably, the cloud-based fusion analysis platform is further configured as follows: Based on the fusion analysis results, a comprehensive crop health index is quantified and generated. The confidence level of the current assessment results is verified by combining the varietal characteristics and planting habits in historical production data.

[0010] Preferably, the decision push module specifically includes: The visualization submodule is used to display digital maps of farmland, health heat maps, and environmental parameter curves on the agricultural information platform; The scheme generation submodule is used to automatically generate differentiated fertilization schemes, irrigation schemes, and pest and disease control schemes based on crop health dynamic assessment reports. The instruction issuing submodule is used to convert decision-making schemes into control instructions that can be executed by agricultural machinery and equipment.

[0011] Preferably, the decision-making push module also incorporates market data. When generating decision-making schemes, it analyzes the input-output ratio in conjunction with the current agricultural product market price trends and recommends the management strategy with the best economic benefits.

[0012] Preferably, the system further includes a dynamic partitioning module, which is used to divide the field area into several operation plots with different management priorities according to the comprehensive crop health index, so as to realize variable operation.

[0013] Preferably, the system also includes a feedback optimization module, which is used to collect crop response data after the implementation of the decision plan, and transmit the response data back to the cloud-based fusion analysis platform to update and iteratively optimize the parameters of the machine learning model online.

[0014] The beneficial effects of this invention are: 1. By integrating multimodal data such as remote sensing, near-ground data, soil data, and historical market data, a comprehensive crop perception system was constructed, effectively solving the diagnostic blind spots caused by single data sources. Spatiotemporal alignment technology was used to eliminate the scale differences in time and space between heterogeneous data, enabling precise correlation analysis between underground soil environment and above-ground crop growth, significantly improving the accuracy of identifying stress states such as pests and diseases, and water and fertilizer deficiencies.

[0015] 2. The system combines historical production and market data for comprehensive analysis, focusing not only on the crop's growth status but also incorporating economic benefit assessment dimensions, making the generated decision-making solutions more scientific and practical. Coupled with visual displays and automated command issuance, it achieves closed-loop control from data collection to intelligent decision-making and precise execution, truly enabling on-demand input, reducing agricultural production costs and mitigating the environmental pollution risks associated with excessive fertilizer and pesticide use. Attached Figure Description

[0016] Figure 1 This is an overall system diagram of the present invention.

[0017] Figure 2 This is a schematic diagram of the structure of the multimodal data acquisition module of the present invention.

[0018] Figure 3 This is a schematic diagram of the edge preprocessing module of the present invention.

[0019] Figure 4 This is a schematic diagram of the cloud-based fusion analysis platform of the present invention.

[0020] Figure 5 This is a schematic diagram of the decision push module structure of the present invention. Detailed Implementation

[0021] The present invention will be further described below with reference to embodiments.

[0022] Example 1: Water stress and precision irrigation during the greening stage of winter wheat This embodiment was applied to a large-scale winter wheat planting base in the North China Plain, aiming to solve the problems of increased water evaporation and hidden drought caused by rising temperatures during the greening period.

[0023] 1. Data Acquisition and Fusion During a clear, cloudless period in mid-March, the system initiated multimodal data acquisition. The remote sensing unit controlled the drone to acquire RGB and multispectral images at a resolution of 0.5 meters, focusing on extracting the Normalized Difference Vegetation Index (NDVI) and Photochemical Vegetation Index (PRI). Simultaneously, the near-ground sensing unit recorded air temperature, relative humidity, and wind speed every 30 minutes; the soil detection unit monitored soil volumetric water content, nitrate nitrogen content, and pH value at the 0-20cm and 20-40cm soil layers, respectively. In addition, the system retrieved yield distribution maps and basal fertilizer application records at the time of sowing for the past three years as historical data for this plot.

[0024] 2. Analysis and Evaluation The edge preprocessing module performs spatial interpolation matching between UAV imagery and soil detection points. The cloud-based fusion analysis platform loads a dedicated model for winter wheat's greening-up stage to analyze the data. The analysis results show that the NDVI value in the northwest corner of the plot is within the normal range, but the deep soil moisture content is only 45% (below the threshold of 60%). Combined with recent high-temperature weather forecasts, the system determines that this area has a "hidden drought" risk. The system generates a crop health dynamic assessment report, marking this area as a "high-risk water-deficient area" and the remaining areas as "normal."

[0025] 3. Decision-making and execution The decision-making module, combining the current wheat market price (2400 yuan / ton) and irrigation costs, calculated that without timely irrigation, the area would suffer a yield reduction of approximately 50 kg per mu, resulting in economic losses exceeding the irrigation investment. The system automatically generated a variable irrigation prescription map, setting the irrigation quota at 30 cubic meters per mu for high-risk areas and 15 cubic meters per mu for normal areas. The instruction delivery submodule sent the prescription map to the integrated water and fertilizer equipment in the field. The equipment automatically adjusted the solenoid valve opening duration and water pump frequency to execute precision irrigation. One week later, the system collected data again, confirming that crop growth in the high-risk areas had returned to normal.

[0026] Example 2: Early warning and green control of diseases and pests during flowering of greenhouse tomatoes This embodiment was applied to a greenhouse tomato growing area in southern China, focusing on early warning and control of gray mold and leaf mold during the flowering period.

[0027] 1. Data Acquisition and Fusion During the initial flowering stage of tomatoes, the system continuously monitors the plants. The near-ground sensing unit focuses on collecting data on air humidity (maintaining above 85% is prone to disease), diurnal temperature range, and CO2 concentration within the greenhouse. The multimodal data acquisition module uses hyperspectral imaging technology to scan leaves, capturing subtle spectral changes caused by early disease (such as shifts in the red edge position). The soil detection unit monitors the EC value (electrical conductivity) of the rhizosphere soil to prevent decreased plant resistance due to excessive salinity. Simultaneously, the system incorporates historical data, including the variety's resistance level to gray mold and the current pest and disease situation in surrounding production areas.

[0028] 2. Analysis and Evaluation The cloud-based fusion analysis platform used machine learning algorithms to perform fusion analysis on the above data. The model found that the spectral reflectance of the leaves in the southeast corner of the greenhouse fluctuated abnormally around 760nm, and the nighttime humidity exceeded 90% for three consecutive days. Combined with the historical data showing that this variety was highly susceptible to disease under these conditions, the system determined that the probability of gray mold outbreak in this area was over 80%. The system generated a health assessment report, identifying the hotspots of disease occurrence and predicting that without intervention, the disease would spread to the entire greenhouse within 5 days.

[0029] 3. Decision-making and execution Based on the assessment results and market data (price differences and residue standards between biological and chemical pesticides), the decision-making module recommends a strategy of "primarily biological control, supplemented by localized emergency chemical treatments." The system generates precise decision-making plans: for the high-risk area in the southeast corner, it recommends targeted spraying with a specified concentration of biological fungicide; for other areas, it suggests strengthening ventilation and dehumidification. The plan is pushed to farmers' mobile phones through the agricultural information platform and synchronized to the task list of the plant protection drone. After farmer confirmation, the drone performs targeted spraying. Subsequent tracking data shows that the disease was effectively controlled, and pesticide residue tests on the fruit met standards, enhancing the product's market competitiveness.

[0030] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A field crop health dynamic assessment system based on multimodal agricultural data fusion, characterized in that, include: The multimodal data acquisition module is used to simultaneously acquire remote sensing image data, near-ground environmental data, multidimensional soil data, and historical production data of the target area; An edge preprocessing module, connected to the multimodal data acquisition module, is used to perform spatiotemporal alignment, cleaning, and feature extraction on the multimodal data; The cloud-based fusion analysis platform communicates with the edge preprocessing module to construct a multi-source agricultural big data set and uses machine learning algorithms to intelligently analyze crop growth status and pest and disease risks, generating a dynamic crop health assessment report. The decision push module is used to visualize the generated accurate decision-making solutions and push them to user terminals or agricultural machinery equipment.

2. The field crop health dynamic assessment system based on multimodal agricultural data fusion according to claim 1, characterized in that, The multimodal data acquisition module includes: The remote sensing unit is configured as a drone or satellite remote sensing device equipped with a multispectral camera to acquire information on the spatial distribution of farmland and spectral data on crop growth. The near-ground sensing unit is configured as an IoT base station deployed in the field to collect data on air temperature and humidity, light intensity, rainfall, and CO2 concentration. The soil detection unit is configured as a composite probe embedded in the root layer to detect soil volumetric water content, nutrient content, pH value and electrical conductivity; Data storage unit, used to store historical agricultural production records and market data.

3. The field crop health dynamic assessment system based on multimodal agricultural data fusion according to claim 2, characterized in that, The edge preprocessing module includes: The spatiotemporal alignment submodule is used to assign a unified time label to heterogeneous data from different sources and map them to the same geographic coordinate system; The data cleaning submodule is used to remove outliers and missing values, and to correct sensor drift data. The feature extraction submodule is used to extract vegetation index features, environmental stress features, and soil fertility features from the raw data.

4. The field crop health dynamic assessment system based on multimodal agricultural data fusion according to claim 1, characterized in that, The cloud-based fusion analysis platform includes a data fusion layer and a model inference layer; The data fusion layer is used to perform multi-source data fusion modeling on the extracted features and construct a multi-source agricultural big data set that includes spatiotemporal dimensions, environmental dimensions, and growth dimensions. The model inference layer has multiple built-in machine learning models for different crop growth cycles, which are used to analyze the current growth status of crops, predict the risk of pest and disease outbreaks, and yield trends.

5. A field crop health dynamic assessment system based on multimodal agricultural data fusion according to claim 4, characterized in that, The cloud-based fusion analysis platform is also configured as follows: Based on the fusion analysis results, a comprehensive crop health index is quantified and generated. The confidence level of the current assessment results is verified by combining the varietal characteristics and planting habits in historical production data.

6. The field crop health dynamic assessment system based on multimodal agricultural data fusion according to claim 1, characterized in that, The decision-making push module specifically includes: The visualization submodule is used to display digital maps of farmland, health heat maps, and environmental parameter curves on the agricultural information platform; The scheme generation submodule is used to automatically generate differentiated fertilization schemes, irrigation schemes, and pest and disease control schemes based on crop health dynamic assessment reports. The instruction issuing submodule is used to convert decision-making schemes into control instructions that can be executed by agricultural machinery and equipment.

7. A field crop health dynamic assessment system based on multimodal agricultural data fusion according to claim 6, characterized in that, The decision-making recommendation module also incorporates market data. When generating decision-making plans, it analyzes the input-output ratio in conjunction with the current market price trends of agricultural products and recommends the management strategy with the best economic benefits.

8. The field crop health dynamic assessment system based on multimodal agricultural data fusion according to claim 1, characterized in that, The system also includes a dynamic partitioning module, which is used to divide the field area into several operation plots with different management priorities according to the comprehensive crop health index, so as to realize variable operation.

9. A field crop health dynamic assessment system based on multimodal agricultural data fusion according to claim 1, characterized in that, The system also includes a feedback optimization module, which collects crop response data after the implementation of the decision-making scheme and sends the response data back to the cloud-based fusion analysis platform to update and iteratively optimize the parameters of the machine learning model online.

Images