Four-color code assignment method based on dual-temporal growth monitoring result of unmanned aerial vehicle

Abstract

The invention relates to the technical field of agricultural management, in particular to a four-color code assignment method based on a dual-temporal growth monitoring result of an unmanned aerial vehicle. The method comprises the following steps: performing data acquisition by using an unmanned aerial vehicle to obtain acquired image data; performing data preprocessing on the acquired image data to obtain a preprocessed image; the preprocessing comprises basic image generation processing, radiation unification and interference extraction processing and analysis unit and vegetation index calculation processing; performing growth monitoring and health judgment processing based on the preprocessed image to obtain a single-phase growth monitoring image; the growth vigor monitoring and health judgment processing comprises normalized smoothing processing; comparing health binarization results of two adjacent stages based on a single-stage growth monitoring image, interpreting state change of each analysis unit in the field, and outputting comparison data based on four-color codes; and outputting work order distribution data based on the comparison data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of smart agriculture technology, and particularly relates to a four-color code assignment method based on double-time-phase growth monitoring results of an unmanned aerial vehicle. BACKGROUND

[0002] With the rapid development of smart agriculture technology, how to efficiently and accurately monitor the growth of crops in a large area has become a major challenge for modern agricultural management. The traditional method is manual field investigation, which has problems such as low efficiency, strong subjectivity, limited statistical area, and cannot meet the needs of large-scale farmland production and dynamic response management of rapid crop growth.

[0003] In recent years, unmanned aerial vehicle (UAV) technology has developed rapidly. Unmanned aerial vehicle remote sensing technology provides a revolutionary data acquisition means for farmland monitoring with its high resolution and high mobility. By carrying multi-spectral or hyperspectral sensors, unmanned aerial vehicles can monitor multiple indicators reflecting crop growth conditions, thereby assessing crop growth. However, most current unmanned aerial vehicles still remain in single-time-phase static growth analysis, which is difficult to dynamically track crop growth trends and cannot be used for precise agricultural scheduling and field investigation. In order to solve this technical problem, a four-color code assignment method based on double-time-phase growth monitoring results of an unmanned aerial vehicle is proposed. SUMMARY

[0004] In order to solve the above technical problems in the prior art, the present application provides a four-color code assignment method based on double-time-phase growth monitoring results of an unmanned aerial vehicle.

[0005] To achieve the above purpose, the embodiments of the present application provide the following technical solutions: In one embodiment provided by the present application, a four-color code assignment method based on double-time-phase growth monitoring results of an unmanned aerial vehicle is provided, which comprises the following steps: Data acquisition is performed using an unmanned aerial vehicle to obtain collected image data; Data preprocessing is performed on the collected image data to obtain preprocessed images; the preprocessing includes basic image generation processing, radiation consistency and interference extraction processing, and analysis unit and vegetation index calculation processing; Based on the preprocessed images, growth monitoring and health determination processing is performed to obtain single-period growth monitoring images; the growth monitoring and health determination processing includes normalization smoothing processing; Based on the single-period growth monitoring images, the health binary results of adjacent two periods are compared, the state change of each analysis unit in the field is interpreted, and four-color code-based comparison data is outputted; Based on the comparison data, work order dispatching data is outputted.

[0006] As a further scheme of the present application, the base image generation processing comprises image de-distortion processing, aerial triangulation processing and orthoimage generation processing.

[0007] As a further scheme of the present application, the image de-distortion processing performs geometric correction on each image in the input collected image data, eliminates radial and tangential distortion caused by lens optical characteristics, and outputs the corrected collected image data.

[0008] As a further scheme of the present application, the aerial triangulation processing is used to automatically identify homonymic points between images in the corrected collected image data by a motion recovery structure algorithm, solve the exterior orientation elements (position and posture) of each image, and generate a sparse point cloud.

[0009] As a further scheme of the present application, the orthoimage generation processing uses DSM to grid the sparse point cloud, takes the highest point elevation in each grid, generates a DSM representing the crop top canopy surface, ortho-projects each non-distorted image onto a plane based on the DSM, and performs color balancing and seamless mosaicking, and outputs an orthoimage map.

[0010] As a further scheme of the present application, the radiation unification and interference extraction processing comprises radiation unification processing and mask extraction processing.

[0011] As a further scheme of the present application, the normalization smoothing processing comprises normalization processing, smoothing processing and statistical unit aggregation processing.

[0012] As a further scheme of the present application, based on the preprocessed images, a vigor monitoring and health determination processing is performed to obtain a single-period vigor monitoring image; the vigor monitoring and health determination processing comprises normalization smoothing processing, which comprises: Based on the preprocessed images, an Otsu / clustering model or a supervised model is used to perform a vigor monitoring and health determination processing to obtain a single-period vigor monitoring image.

[0013] As a further scheme of the present application, the comparison data based on the four-color code comprises raster data, vector data and statistical report.

[0014] As a further scheme of the present application, based on the comparison data, work order dispatching data is outputted, which comprises: The comparison data is sequentially subjected to warehousing, mask processing, index calculation, plot statistics, health determination, code color generation, result publishing and task dispatching, etc., to output the work order dispatching data.

[0015] The technical scheme provided by the present application has the following beneficial effects: 1. This invention is based on high-resolution remote sensing images of farmland acquired by unmanned aerial vehicles (UAVs), and develops a four-color code assignment method based on the UAV's dual-temporal growth monitoring results. This method periodically monitors crop growth and interprets state changes in adjacent two-phase (dual-temporal) results, forming a four-color code (green / yellow / purple / red) visualization and early warning closed loop for precise agricultural scheduling, field inspection, and management assessment.

[0016] 2. By comparing the monitoring results of two adjacent periods, the status changes of each analysis unit are summarized into four typical situations: maintaining health, deteriorating health, improving from poor, and persistently poor. These are visually identified using four color codes: green, yellow, purple, and red. This overcomes the limitations of traditional single-phase static analysis, enabling managers to clearly understand which areas in the farmland are stable, which areas have new problems, and which areas are recovering. This greatly enhances the insight into crop growth dynamics and achieves accurate perception and intuitive expression of dynamic changes in growth.

[0017] 3. This method integrates UAV remote sensing, image processing, threshold determination, and rule mapping into an automated process, replacing inefficient and subjective manual inspections. The system can automatically generate four-color code maps and statistical reports with geographic information, and can directly generate agricultural work orders, accurately pushing early warning information to responsible persons. This significantly shortens the response time from problem discovery to initiation of action, improves the efficiency and accuracy of management decisions, and enhances the automation level and decision-making efficiency of farmland monitoring.

[0018] 4. This method goes beyond status diagnosis; it guides field verification and agricultural operations through work order dispatch and feeds the verification results back to the system to optimize health assessment thresholds. This closed-loop design enables the system to learn and evolve independently. As data accumulates, its accuracy and adaptability to local planting conditions will continuously improve, forming a sustainable and optimized smart agriculture management cycle and constructing a management closed loop of monitoring, analysis, decision-making, execution, and optimization.

[0019] 5. This method employs quality control measures such as radiation homogenization, interference masking, noise reduction analysis based on statistical values, and dynamic thresholds for crop and growth stage management. These measures ensure the stability and comparability of analysis results under different weather and light conditions. Furthermore, the core four-color code interpretation logic is independent of specific sensor types and vegetation indices, is compatible with multispectral and RGB data, and allows for adjustment of the analysis unit granularity as needed, demonstrating good applicability and scalability.

[0020] These or other aspects of the invention will become more apparent from the following description of embodiments. It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of a four-color code assignment method based on the dual-temporal-phase growth monitoring results of an unmanned aerial vehicle (UAV) according to an embodiment of the present invention; Figure 2 This is a flowchart of the preprocessing and post-preprocessing image processing in the four-color code assignment method based on the dual-temporal-phase growth monitoring results of UAVs in an embodiment of the present invention. Figure 3 This is a schematic diagram of the four-color code rules for the two-temporal four-color code assignment method based on the two-temporal growth potential monitoring results of UAV proposed in this invention; Figure 4 This is a schematic representation of the color specification for the four-color code assignment method based on the dual-temporal-phase growth monitoring results of UAV proposed in this invention; Figure 5 This is an example diagram of the four-color code visualization based on the dual-temporal growth monitoring results of UAV 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0025] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0026] UAV: Unmanned Aerial Vehicle. This solution requires the UAV to be equipped with visible light or multispectral drones and have high-precision positioning capabilities.

[0027] DOM / Ortho: Orthophotos are images with accurate geographic coordinates that have undergone geometric correction, radiometric correction, and mosaicking to eliminate distortions caused by sensor tilt, terrain undulation, and central projection.

[0028] GSD: Ground Sampling Distance, which represents the size of a pixel on the actual ground.

[0029] GCP: Ground Control Point, a point located on the ground with precise absolute coordinates.

[0030] DSM / DEM / CHM: Digital Surface / Terrain / Crowne Height Model.

[0031] VI: Vegetation index, calculated from different bands of multispectral imagery through linear or nonlinear combinations, used to enhance and quantify vegetation characteristics (e.g., NDVI, NDRE, NGRDI, ExG).

[0032] NDVI: Normalized Difference Vegetation Index.

[0033] NDRE: Normalized Difference Red Edge Index.

[0034] NGRDI: Normalized Green-Red Difference Index.

[0035] ExG: Excess Green Index GNDVI: Green Normalized Difference Vegetation Index SAVI: Soil Adjusted Vegetation Index FCover: Fraction of Vegetation Cover ROI / Plot: A plot of land / job subplot, the basic unit for analysis.

[0036] Healthy / Unhealthy: A binary classification result based on growth index thresholds or model determination.

[0037] Specifically, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0038] Please see Figure 1 , Figure 1 This is a flowchart of a four-color code assignment method based on the dual-temporal-phase growth potential monitoring results of an unmanned aerial vehicle (UAV) according to an embodiment of the present invention, as shown below. Figure 1 As shown, the four-color code assignment method based on the dual-temporal growth monitoring results of UAV includes steps S10 to S50.

[0039] S10. Use drones to collect data to obtain image data; In an embodiment of the present invention, the drone can acquire images using an RGB visible light image acquisition device and a multispectral image acquisition device.

[0040] RGB visible light image acquisition devices should have a resolution of ≥20MP (20 million pixels). For multispectral image acquisition devices, it is recommended to configure sensors that include four bands: red, green, near-infrared, and red-edge.

[0041] RGB visible light image acquisition devices are used to generate high-precision orthophotos for visual interpretation, texture analysis, and calculation of visible light vegetation indices. They are an essential basic payload. They are used to calculate key vegetation indices sensitive to crop physiological parameters, such as NDVI and NDRE. Multispectral image acquisition devices are used for image acquisition.

[0042] In embodiments of the present invention, the UAV can fly at an altitude of 80–120 m. This altitude range represents the optimal balance between accuracy and efficiency. Too low an altitude results in low efficiency, while too high an altitude leads to excessive GSD and loss of detail.

[0043] like Figure 2 As shown, S20, the acquired image data is preprocessed to obtain a preprocessed image; the preprocessing includes basic image generation processing, radiometric homogenization and interference removal processing, and analysis unit and vegetation index calculation processing.

[0044] The basic image generation process includes image distortion correction, aerial triangulation encryption, and orthophoto generation.

[0045] In an embodiment of the present invention, image distortion correction processing performs geometric correction on each image in the input acquired image data to eliminate radial and tangential distortion caused by lens optical characteristics and outputs corrected acquired image data.

[0046] In an embodiment of the present invention, the aerial triangulation encryption process is used to automatically identify corresponding points between images in the corrected acquired image data through the motion recovery structure algorithm, calculate the exterior orientation elements (position and attitude) of each image, and generate a sparse point cloud.

[0047] In an embodiment of the present invention, the orthophoto generation process uses DSM to grid the sparse point cloud, takes the elevation of the highest point in each grid, generates a DSM representing the top surface of the crop canopy, and based on the DSM, projects each distortion-free image onto the plane, performs color equalization and seamless mosaicking, and outputs an orthophoto map.

[0048] In embodiments of the present invention, the radiation homogenization and interference extraction process includes radiation homogenization processing and mask extraction processing.

[0049] In embodiments of the present invention, radiometric normalization is used to sequentially perform relative radiometric normalization and statistical normalization on orthophoto maps to obtain a multi-period DOM dataset with high radiometric consistency.

[0050] Relative radiation normalization: Select one period as the reference period, find spectrally stable pseudo-invariant feature points in other periods, establish a linear or nonlinear transformation model, and unify the radiation level of the images in each period to the reference period.

[0051] Based on statistical normalization: if dual-phase data is unavailable, histogram matching or standard deviation stretching can be performed on single-phase images to make their statistical characteristics more consistent.

[0052] In an embodiment of the present invention, the mask extraction process is used to identify clouds and shadows, water bodies and bare land / soil in a multi-period DOM dataset, and output a composite mask file that identifies all analysis areas that need to be excluded, i.e., output the multi-period DOM dataset after mask extraction processing.

[0053] Clouds and Shadows: Utilizing the high reflectivity of clouds and the low brightness of shadows, detection is performed using thresholding or machine learning models to generate masks.

[0054] Water bodies: Identification is achieved by utilizing the strong absorption characteristics (low values) of the near-infrared band on water bodies, combined with the visible light band.

[0055] Bare land / soil: Use vegetation indices (such as NDVI) and set thresholds (such as NDVI < 0.2) to distinguish between them.

[0056] In embodiments of the present invention, the analysis unit and vegetation index calculation processing include analysis unit processing and vegetation index calculation processing.

[0057] The analysis unit processes both raster grid processing and vector plot processing.

[0058] Raster grid processing: Dividing the entire farmland into regular grids, it features fine-grained analysis, precise change location, and is not limited by management boundaries, making it easy to automate. It is recommended as the preferred unit for four-color code change interpretation.

[0059] Vector plot processing: Using actual field boundaries (ROI / Plot) as the unit of analysis, it conforms to management practices, is easy to link with agricultural records, and is suitable for overall crop growth assessment and report generation at the plot level.

[0060] Vegetation index calculation processing is used to calculate vegetation indices based on the analysis data obtained from the analysis unit processing and the multi-period DOM dataset after mask extraction processing, in order to obtain preprocessed images.

[0061] It should be noted that, when multispectral data is available, the vegetation index calculation process prioritizes NDVI and NDRE, which are more sensitive to chlorophyll and biomass. NDVI is suitable for the early and mid-growth stages, while NDRE performs better in the mid-to-late growth stages (high canopy coverage).

[0062] It should be noted that when only RGB data is available, the vegetation index calculation process uses visible light vegetation indices such as NGRDI or ExG as effective alternatives. However, it should be observed that these indices are more sensitive to light intensity and soil background.

[0063] Continue to refer to Figure 2 As shown, in step S30, growth monitoring and health assessment are performed based on the preprocessed images to obtain single-period growth monitoring images; the growth monitoring and health assessment includes normalization smoothing.

[0064] In embodiments of the present invention, the normalization smoothing process includes normalization processing, smoothing processing, and statistical unit aggregation processing.

[0065] The normalization process involves simultaneous z-score / Min-Max normalization. The former is suitable for data distributions that are close to normal and can effectively highlight anomalies relative to the average level. The latter scales the values ​​to the [0,1] interval. It is more intuitive but sensitive to outliers.

[0066] The smoothing process involves spatial 3×3 or 5×5 median / Gaussian filtering for noise reduction. The former effectively removes noise while better preserving edge information, while the latter performs a weighted average, making the image smoother but potentially causing edge blurring. Median filtering is preferred.

[0067] The statistical unit aggregates the following data: statistics within the plot (mean, P25, P50, P75, std).

[0068] In an embodiment of the present invention, S30, based on the preprocessed image, growth monitoring and health assessment processing is performed to obtain a single-period growth monitoring image; the growth monitoring and health assessment processing includes normalization smoothing processing, including: Based on the preprocessed images, growth monitoring and health assessment are performed using the Otsu / clustering model or a supervised model to obtain single-period growth monitoring images.

[0069] The health thresholds for the Otsu / clustering model and the supervised model can be empirical thresholds or quantile thresholds.

[0070] Empirical thresholds: For example, NDVI ≥ 0.55 indicates good health; Quantile threshold: For the same crop and the same growth period, P50–P70 is the health threshold.

[0071] S40. Based on single-period growth monitoring images, compare the health binarization results of two adjacent periods, interpret the status changes of each analysis unit in the field, and output comparison data based on four-color codes.

[0072] The comparison data based on the four-color code includes raster data, vector data, and statistical reports.

[0073] Raster data: Generate a four-color raster map (GeoTIFF) with the same resolution as the DOM, which is convenient for overlaying and viewing in the GIS platform.

[0074] Vector data: Output each analysis unit (grid) and its four-color code attributes as a vector file (such as GeoJSON or Shapefile), which should include fields such as color assignment, health status of the previous and current periods, and analysis unit.

[0075] Statistical Reports: Automatically generates statistical reports, including the area, percentage, and spatial distribution list of each color code.

[0076] S50. Based on the comparison data, generate corresponding agricultural management work orders; among them, generate field inspection work orders for analysis units assigned yellow and red codes, and generate review work orders for analysis units assigned purple codes.

[0077] S50, based on the comparison data, generates corresponding agricultural management work orders, including: The data is processed sequentially, including data entry, masking, indicator calculation, land parcel statistics, health assessment, code generation, result release, and task assignment, to output work order assignment data.

[0078] The data processing includes inputting raster data, vector data, and statistical reports from adjacent periods, performing automated data verification and unified coordinate system conversion; establishing a versioned management system for core data assets (DOM, index, land parcel) to ensure that every analysis process is traceable, and outputting versioned multi-period standard datasets stored in the spatiotemporal database.

[0079] Masking involves inputting a standard dataset and predefined masking rules into the database, calling a unified masking service, and sequentially applying static masks (roads, buildings) and dynamic masks (clouds, shadows, bare ground) to generate a 0-composite "effective analysis area" mask. This eliminates interference for subsequent analysis and outputs a mask file that identifies pure vegetation pixels.

[0080] The index calculation is used to automatically route to the corresponding computing engine based on the data type (multispectral / RGB), and to calculate a series of growth indicators such as NDVI, NDRE, CHM, and texture features in parallel, which greatly improves processing efficiency.

[0081] Plot statistics are used to calculate multiple statistical measures such as the mean, P25, P75, and standard deviation of the effective pixels within each plot, so as to comprehensively depict the overall level and uniformity of growth within the plot.

[0082] The health assessment function uses the plot ID and crop growth stage to call preset thresholds from a dynamic threshold table, compares the plot's representative statistical value (e.g., P75) with the threshold, and automatically determines its health status (1 / 0). Plots with poor data quality are marked and allowed for error correction.

[0083] The code generation function is used to generate a four-color code for each plot based on the health status of the plots in the two phases, applying the core state machine rules, and outputting vector layers and thematic images with four-color code attributes.

[0084] Results are published using four-color vector layers and statistical reports, and the results are released through multiple channels: published as WebGIS map services for online interaction, generated as PDF briefings, provided as data services via API, and pushed as early warning summaries to mobile devices. The results are standardized monitoring results accessible via Web, API, reports, and mobile devices.

[0085] Task assignment is used to automatically create agricultural work orders based on the four-color code results and according to the four-color code business rules. It generates field inspection work orders for yellow / red code areas and review work orders for purple code areas, and pushes them to the responsible persons through the message system. The structured list of pending work orders is then entered into the task flow system.

[0086] Step S5 of the present invention may further include step S6: receiving field verification feedback data for agricultural management work orders, and optimizing and iterating the health determination threshold based on the feedback data.

[0087] The color standard definition for the four-color code is as follows: green code corresponds to HEX color value #00B050, yellow code corresponds to #FFD966, purple code corresponds to #7030A0, and red code corresponds to #C00000.

[0088] Examples of specific implementation steps: Step 1: Data Collection; Based on the fact that the corn is currently in the jointing stage of growth, this flight mission is set as a dual-temporal monitoring corresponding to the previous cycle (10 days ago). The flight route is planned through flight control software to ensure that the heading overlap is ≥80%, the lateral overlap is ≥75%, the relative flight altitude is 100 meters, and the expected ground resolution (GSD) is 3 centimeters. The operation was conducted under clear, windless weather conditions with a solar altitude angle greater than 45 degrees. The drone was equipped with a 20-megapixel RGB camera and a four-channel multispectral camera that included red, green, red-edge, and near-infrared bands for simultaneous data acquisition. Before the flight, six ground control points (GCPs) were set up around the field and in its center, and their precise WGS84 coordinates were obtained using an RTK measuring instrument. After the flight is completed, the original image data, POS data and GCP coordinate files of the two phases (referred to as phase T-1 and phase T respectively) are uploaded to the designated storage directory of the data processing server.

[0089] Step 2: Dual-phase data processing and growth indicator extraction; The image data needs to be geometrically corrected and radiometrically consistent. Aerial triangulation is performed on the multispectral and RGB images of T-1 and T periods respectively. GCP coordinates are imported to enhance absolute positioning accuracy. Based on the generated dense point cloud and digital surface model (DSM), multispectral orthophotos (DOM) and RGB orthophotos for each time phase are generated. The T-1 period DOM is set as the reference benchmark. Pseudo-invariant feature points such as bare soil and roads are automatically selected on the T period image. The radiometric normalization coefficient is calculated, and the T period DOM is radiometrically corrected so that the spectral response of the two periods of images tends to be consistent. Cloud, shadow, and water body detection were performed on the two phases of imagery, and bare soil masks were generated using an NDVI threshold of <0.2. All masks were then merged into a single composite invalid region mask file. Import the pre-digitized field boundary vector file (each field has a unique plot_id and is associated with the maize-jointing stage attribute); Calculate at least one vegetation index to characterize crop growth: For multispectral DOM, NDVI (Normalized Differential Vegetation Index) and NDRE (Normalized Differential Red Edge Index) can be calculated for each pixel. For RGB DOM, ExG (Extra Green Index) can be calculated as a reference. For each field (analysis unit), an invalid region mask is applied to retain only pure vegetation pixels. If the proportion of valid pixels in a field is less than 65%, the field is marked as insufficient data (gray code). For valid fields, the median (P50) of the NDVI values ​​of all valid pixels in the field is calculated as the statistical values ​​V_t-1 and V_t representing the growth level of the field.

[0090] Step 3: Health status assessment and four-color code generation; like Figure 2 As shown, for each field, its statistical value is compared with a threshold: The threshold Th is an adaptive threshold determined based on historical sample data or statistical distribution of similar plots, used to characterize the normal growth range of the corresponding crop under the same growth period.

[0091] If V_t-1>=Th, then the health status in period T-1 is determined to be H_t-1=1 (healthy); otherwise, H_t-1=0 (unhealthy).

[0092] Similarly, if V_t>=Th, then the health status in period T is determined to be H_t=1; otherwise, H_t=0. according to Figure 2 The four-color code assignment rule shown performs logical judgments on each field: If H_t-1==1 and H_t==1, then the field is assigned a green code, indicating that the crops are growing healthily.

[0093] If H_t-1==1 and H_t==0, then the field is assigned a yellow code, indicating that the crop condition has deteriorated and a warning is issued.

[0094] If H_t-1==0 and H_t==1, then the field is assigned a purple code, which means that the growth has improved from poor to good, indicating that the recovery or measures have been effective.

[0095] If H_t-1==0 and H_t==0, then the field is assigned a red code, indicating that the growth is still unhealthy and needs close attention.

[0096] The fourth step is to publish the results; like Figure 3 As shown, the system automatically generates a four-color thematic map. On the WebGIS platform, each field is rendered according to its color code (green: #00B050, yellow: #FFD966, purple: #7030A0, red: #C00000), and supports clicking to query details (such as plot_id, V_t-1, V_t, H_t-1, H_t, and the meaning of the color code). Statistical reports are also generated, listing the number, area, and percentage of fields with each color code. Create field inspection work orders for all yellow and red code fields. The work orders include the field location, number, and changes (new / ongoing) and are pushed to the mobile APP of the relevant person in charge through the message center. Create a work order for reviewing the restoration effect of the purple code field and prompt the person in charge to confirm it; Field managers receive work orders via a mobile app, navigate to designated fields for on-site inspection, record the causes of problems such as pests and diseases, water shortage, and fertilizer damage, and upload photos. After the inspection is completed, they submit a work order completion report in the app. All feedback results from field verifications, including actual conditions and corresponding image index values, are stored in the database. The system regularly performs statistical analysis on crop type, growth stage, and actual growth data every month. It can automatically or semi-automatically calibrate and update the thresholds in the dynamic threshold table, making health assessments more consistent with actual field conditions and forming a closed-loop management system of monitoring, decision-making, action, feedback, and optimization.

[0097] Specific measurement requirements: In the actual measurement process, ground control is necessary to eliminate projection distortion and sensor error and achieve accurate pixel-level registration between multiple data periods. For ground control point (GCP) setup, the density should be set at 5–8 GCPs per 20–40 hectares; the locations should be evenly distributed around the perimeter and center of the survey area, ensuring that each GCP is clearly visible on at least 2–3 images, and should be placed on permanent features such as field ridges and road intersections; when management operations require absolute accuracy <10 cm (such as precision variable fertilization, autonomous driving navigation), GCPs must be deployed, and their precise coordinates should be acquired using RTK or PPK mode; Regarding the timing and cycle planning, a timing interval of 7-14 days is recommended (which can be slightly adjusted depending on the crop and management rhythm). This cycle can effectively monitor the changes in the key growth period of crops, as well as the typical stress development process caused by water, fertilizer, pests and diseases, and avoid missing important change signals due to excessively long intervals. During key growth periods, a shorter cycle is recommended; under severe weather conditions, the cycle can be extended; and in the case of pest infestations, the cycle can be shortened to 3-6 days.

[0098] In terms of data preprocessing, this process transforms the raw UAV imagery into standardized, reliable data products that can be directly used for crop growth analysis and change interpretation. All steps must be completed on professional remote sensing software or a self-developed algorithm platform. During the basic image generation process, each image is geometrically corrected using camera factory calibration parameters (focal length, principal point, lens distortion coefficient) after input from UAV images. This eliminates radial and tangential distortions caused by lens optical characteristics. The corrected images are then output. The corrected images are then processed using a motion recovery structure algorithm to automatically identify corresponding points between images, calculate the exterior orientation elements (position and attitude) of each image, and generate a high-precision sparse point cloud. The dense point cloud is then gridded, and the elevation of the highest point within each grid is taken to generate a DSM representing the surface of the crop canopy. Based on the DSM, each distortion-free image is orthophotographed onto a plane, and color equalization and seamless mosaicking are performed to finally obtain an orthophoto map. Regarding radiometric consistency and interference removal, relative radiometric normalization involves selecting one period as a reference period, searching for spectrally stable pseudo-invariant feature points in other periods, establishing linear or nonlinear transformation models, and unifying the radiometric levels of images from each period to the reference period. Statistical normalization involves performing histogram matching or standard deviation stretching on single-period images if dual-phase data is unavailable, making their statistical characteristics more consistent, and finally obtaining a multi-period DOM dataset with high radiometric consistency. Secondly, regarding clouds and shadows, the high reflectivity of clouds and the low brightness of shadows are utilized to detect and generate masks using thresholding or machine learning models; for water bodies, the strong absorption characteristics (low values) of near-infrared bands are used in conjunction with visible light bands for identification; for bare land / soil, vegetation indices (such as NDVI) are used, and thresholds (such as NDVI < 0.2) are set to distinguish them, ultimately resulting in a composite mask file that identifies all analysis areas that need to be excluded. When analyzing units and calculating vegetation indices, the entire farmland is divided into regular grids, which features fine analysis granularity, precise change location, and no restrictions from management boundaries. It is easy to automate and is recommended as the first choice for four-color code change interpretation to form a raster grid. Using actual field boundaries (ROI / Plot) as the analysis unit conforms to management habits, is easy to link with agricultural records, and is suitable for plot-level overall growth assessment and report generation to generate vector plots. Secondly, the raster grid or vector plots are extracted using a mask to obtain multi-period DOM datasets with high radiometric consistency. When multispectral data is available, NDVI and NDRE, which are more sensitive to chlorophyll and biomass, are preferred. NDVI is suitable for the early and middle stages of growth, while NDRE performs better in the middle and late stages of growth (high-coverage canopy). When only RGB data is available, visible vegetation indices such as NGRDI or ExG are used as effective alternatives. However, it should be noted that these indices are more sensitive to light and soil background. Finally, the processed images are calculated, and the monitoring results of each period are binarized into: healthy (1) / unhealthy (0).

[0099] Regarding normalization and smoothing: Normalization methods: simultaneous z-score / Min-Max normalization; the former is suitable for data distributions close to normal and can effectively highlight anomalies relative to the average level. The latter scales values ​​to the [0, 1] interval. It is more intuitive but sensitive to outliers; Smoothing methods: Spatial 3×3 or 5×5 median / Gaussian filtering for noise reduction; the former can effectively remove noise while better preserving edge information, while the latter performs weighted averaging to make the image smoother, but may cause edge blurring. Median filtering is preferred. Statistical unit aggregation: Statistics within the plot (mean, P25, P50, P75, std).

[0100] Regarding health threshold settings and growth monitoring images, choose one or a combination of them: First, input: the processed image; Empirical thresholds: For example, NDVI ≥ 0.55 indicates good health; Quantile threshold: For the same crop and the same growth period, P50–P70 is the health threshold; Otsu / Clustering: Adaptive segmentation based on cell or plot mean; Supervised model: A binary classifier (LR / GBDT / LightGBM) is trained using measured sample plots as labels.

[0101] Engineering recommendations: Prioritize the use of quantile thresholds plus intra-plot statistics to balance robustness and universality; for cross-period comparability, threshold tables need to be maintained within the same plot / same similar growth period dimension.

[0102] Output: Single-period growth monitoring images; the overall process is as follows: Figure 5 As shown.

[0103] The industrialized processing flow is as follows: Data import: two-phase DOM / index raster files, vector plot boundaries, flight mission metadata, etc.; automated verification and unified coordinate system transformation of data; establishment of a versioned management system for core data assets (DOM, index, plot) to ensure traceability of each analysis process; and storage of versioned multi-phase standard datasets in a spatiotemporal database. Masking: The standard dataset and predefined masking rules are input into the database. The unified masking service is called to apply static masks (roads, buildings) and dynamic masks (clouds, shadows, bare land) in sequence to generate a composite effective analysis area mask, which eliminates interference for subsequent analysis and identifies pure vegetation pixels. Indicator Calculation: After masking, the DOM / DSM data is automatically routed to the corresponding calculation engine according to the data type (multispectral / RGB), and a series of growth indicators such as NDVI, NDRE, CHM, and texture features are calculated in parallel, which greatly improves processing efficiency and forms a complete set of indicator raster layers aligned with DOM pixels. Plot statistics: Indicator raster layer, vector plot boundary, take each plot as the statistical unit, calculate multiple statistical measures such as mean, P25, P75, standard deviation and other effective pixels in it, comprehensively describe the overall level and uniformity of growth in the plot, and obtain a detailed data table containing the statistical results of various indicators of each plot. Health determination: Based on the plot ID and crop growth period, the preset threshold in the dynamic threshold table is called, and the representative statistical value of the plot (such as P75) is compared with the threshold to automatically determine its health status (1 / 0). Plots with poor data quality are marked and fault-tolerant. The binary health status (H_{t-1}, H_t) of each plot in the previous period and this period. Code generation: Based on the health status of the plots in the two phases, the core state machine rules are applied to generate a four-color code for each plot, resulting in a vector layer and thematic map image with four-color code attributes; Results Release: The four-color code vector layer and statistical reports will be released through multiple channels: as WebGIS map services for online interaction, as PDF briefings, as data services provided through APIs, and as early warning summaries pushed to mobile devices, providing standardized monitoring results accessible through Web, API, reports, and mobile devices; Task dispatch: Agricultural work orders are automatically created according to the four-color code business rules: For areas with yellow / red codes, based on the status code value, data processing instructions corresponding to the abnormal status are triggered to generate processing task data associated with the abnormal area, generate a structured list of pending work orders, and enter the task flow system. Closed-loop data reporting: Inspectors report on site conditions and photos via mobile devices, forming a data loop; the system learns and optimizes health assessment thresholds regularly based on a large amount of feedback data, achieving continuous improvement in accuracy, and finally forming a closed-loop work order record and an optimized and updated dynamic threshold table.

[0104] In actual testing, if the drone is only equipped with an RGB camera, then in the vegetation index calculation step, ExG or NGRDI is used instead of NDVI. The health judgment threshold needs to be recalibrated based on the visible light index, while the core dual-time comparison and four-color code mapping logic remains unchanged.

[0105] In the health assessment step, in addition to using a fixed value, an adaptive method can also be used for the threshold. For example, for each field, a specific quantile (such as the 75th percentile, P75) of the NDVI values ​​of all pixels within it can be calculated as its representative value V_stat, and then compared with the dynamic threshold. This method is more robust to the case of uneven distribution within the field.

[0106] To ensure the accuracy, reliability, and practicality of crop growth monitoring and change interpretation results in practical applications, the system needs to consider and handle the following key quality control links and boundary conditions; First, select one image as a reference baseline. Then, select spectrally stable pseudo-invariant feature points (such as asphalt roads, bare soil areas, and building roofs) from other image periods to establish linear or nonlinear radiative conversion models. After normalization, the band DN value difference of the pseudo-invariant feature points should be less than 5%. Secondly, the registration error of images from different periods within the same plot should be less than 0.5 pixels. For 5cm GSD images, this means the registration error needs to be less than 2.5cm. Furthermore, once such changes to basic information are detected, they should be treated as new plots to avoid the continuation of incorrect status. Thresholds and statuses need to be reset when plots are rezoned or crops are replaced. Next, the effects of irrigation marks, straw, and leaf spots caused by diseases and pests on the RGB index need to be investigated in conjunction with CHM / texture to suppress false positives; rule-based filters or simple classifiers should be integrated into the health determination module to make decisions based on a combination of VI, CHM, and texture features. Finally, if any plot of land has less than 65% effective vegetation pixels in any period (usually due to severe cloud or shadow cover), it will be marked as GRAY or PENDING and displayed in neutral gray on the four-color thematic map, distinguishing it from the four-color coding system. After acquiring supplementary data, the system will automatically reprocess and update the status of the plot.

[0107] Definition of acceptance and evaluation criteria: To objectively measure the accuracy, stability, and business value of this system in a real production environment, the following quantitative acceptance and evaluation indicators are established. These indicators are unique, and the evaluation methods can be more refined and accurate, but cannot be reversed.

[0108] First, randomly select yellow-coded, red-coded, and some green-coded plots as samples (the total sample size shall not be less than 10% of the total number of plots in the current period). An evaluation team of no fewer than three experienced technicians will conduct independent field surveys and record their conclusions without knowing the system's interpretation results. Finally, compare the system's interpretation results with the manual interpretation results, calculate the proportion of plots with consistent conclusions across all sampled plots, and ensure a spatial consistency rate of ≥85% with the manual field survey conclusions. Secondly, checkpoints are evenly distributed within the area. After acquiring more than two phases of UAV imagery, the coordinates of these checkpoints are measured on the images from the previous and subsequent phases, the coordinate deviations of all checkpoints in the X and Y directions are calculated, and their root mean square error is calculated. The root mean square error (RMSE) for cross-phase registration of the same plot is less than 0.2 meters. Third, multiple sets of image data were collected in the same area under different weather conditions (sunny / partly cloudy / thin clouds) and at different times (morning / noon / afternoon). Using the same set of parameters and thresholds, the growth status of these images was classified separately. The overall difference rate between the classification results (such as the health status of each plot) under different conditions was calculated. The stability fluctuation of the classification results under different light / wind speed conditions was <8%. Fourth, for all yellow and red code work orders generated by the system within the statistical period (such as one week or one month), from the time stamp of the work order generation to the time when the field verification personnel upload on-site photos and records via mobile terminal, calculate the time spent on each work order, and count the number of work orders that are completed within 48 hours and their percentage. Within 48 hours after the yellow / red code warning work order is generated, no less than 80% of the field verification and system backfilling must be completed.

[0109] This invention develops a four-color code assignment method based on high-resolution remote sensing imagery of farmland acquired by unmanned aerial vehicles (UAVs) and dual-temporal crop growth monitoring results. This method periodically monitors crop growth and interprets state changes between adjacent periods (dual-temporal phases), generating a visual four-color code (green / yellow / purple / red) (e.g., ...). Figure 4 It forms a closed loop for early warning and is used for precise agricultural scheduling, field inspection, and management assessment.

[0110] It should be understood that although the above description follows a certain order, these steps are not necessarily executed in that order. Unless otherwise expressly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, some steps in this embodiment may include multiple steps or multiple stages, which are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be performed alternately or in turn with other steps or at least a portion of the steps or stages in other steps.

[0111] It should be understood that, as used herein, the singular form "a" is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, "and / or" refers to any and all possible combinations of one or more of the associatedly listed items. The embodiment numbers disclosed above are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0112] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples. Within the framework of the invention, technical features of the above embodiments or different embodiments can be combined, and many other variations of different aspects of the invention exist, which are not provided in the details for the sake of brevity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

Claims

1. A four-color code assignment method based on a two-phase long-term trend monitoring result of a UAV, characterized in that, The method comprises: Data acquisition is performed by using a UAV to obtain collected image data; Data preprocessing is performed on the collected image data to obtain preprocessed images; the preprocessing comprises basic image generation processing, radiation unification and interference extraction processing, and analysis unit and vegetation index calculation processing; Based on the preprocessed images, a health monitoring and determination processing is performed to obtain single-period health monitoring images; the health monitoring and determination processing comprises normalization smoothing processing; Based on the single-period health monitoring images, a comparison is made between health binary results of adjacent two periods, a state change of each analysis unit in a field is interpreted, and comparison data based on four-color codes are outputted; Based on the comparison data, work order dispatching data are outputted. 2.The four-color code assignment method based on the UAV dual-phase growth potential monitoring result of claim 1, wherein, The basic image generation processing comprises image distortion correction processing, space triangulation encryption processing, and orthographic image generation processing. 3.The four-color code assignment method based on the UAV dual-phase growth potential monitoring result of claim 2, wherein, The image distortion correction processing performs geometric correction on each image in the input collected image data, eliminates radial and tangential distortion caused by lens optical characteristics, and outputs corrected collected image data. 4.The four-color code assignment method based on the UAV dual-phase growth potential monitoring result of claim 3, wherein, The space triangulation encryption processing is used to automatically identify homonymic points between images in the corrected collected image data by a motion recovery structure algorithm, solve the exterior orientation elements (position and posture) of each image, and generate a sparse point cloud. 5.The four-color code assignment method based on the UAV dual-phase growth potential monitoring result of claim 4, wherein, The orthographic image generation processing uses DSM to grid the sparse point cloud, takes the highest point elevation in each grid, generates a DSM representing a crop top surface, orthographically projects each non-distorted image onto a plane based on the DSM, and performs color equalization and seamless splicing to output an orthographic image map. 6.The four-color code assignment method based on the UAV dual-phase growth potential monitoring result of claim 1, wherein, The radiation unification and interference extraction processing comprises radiation unification processing and mask extraction processing.

7. The four-color code assignment method based on the unmanned aerial vehicle two-phase growth potential monitoring result according to claim 1, characterized in that, The normalization smoothing processing comprises normalization processing, smoothing processing, and statistical unit aggregation processing. 8.The four-color code assignment method based on the UAV dual-phase growth potential monitoring result of claim 1, wherein, Based on the preprocessed images, a health monitoring and determination processing is performed to obtain single-period health monitoring images; The health monitoring and determination processing comprises normalization smoothing processing, which comprises: Based on the preprocessed images, an Otsu / clustering model or a supervised model is used to perform a health monitoring and determination processing to obtain single-period health monitoring images. 9.The four-color code assignment method based on the UAV dual-phase growth potential monitoring result of claim 1, wherein, The comparison data based on four-color codes comprise raster data, vector data, and statistical reports. 10.The four-color code assignment method based on the UAV dual-phase growth potential monitoring result of claim 9, wherein, Based on the comparison data, work order dispatching data are outputted, which comprise: The comparison data are sequentially subjected to processing such as warehousing, mask processing, index calculation, field statistics, health determination, code color generation, result release, and task dispatching, to output work order dispatching data.

Images