Sky-ground integrated crop condition monitoring method and system based on intelligent field patrolling

Abstract

The present application relates to a sky-ground integrated agricultural condition monitoring method and system based on intelligent field patrol, and belongs to the technical field of agricultural informatization and unmanned aerial vehicle remote sensing monitoring. An unmanned aerial vehicle image sequence of a target farmland is acquired; spatial registration and geometric correction are performed on the unmanned aerial vehicle image sequence to generate an orthographic image and a digital surface model covering the target farmland; based on elevation mutation information, a planting boundary of the farmland is extracted, and a ridge, road and ditch structure inside the farmland is identified; in combination with a connected domain analysis algorithm, the farmland region is segmented into multiple water and fertilizer management units corresponding to independent spatial vector polygons; based on the spatial vector polygons, a standardized farmland spatial base map is generated, including the boundary, area and spatial topological relationship of each unit. The present application realizes full-process automation from image acquisition to agricultural condition decision, can effectively reduce the intensity of manual field patrol, improve the timeliness and spatial accuracy of farmland anomaly identification, and provide directly executable operation instructions for large-scale precision management.

Description

Technical Field

[0001] This invention relates to a method and system for integrated air-ground agricultural monitoring based on intelligent field patrol, belonging to the field of agricultural informatization and UAV remote sensing monitoring technology. Background Technology

[0002] In existing technologies, some farmers have begun to use drones to acquire orthophotos of farmland and then manually interpret them to identify farmland boundaries, crop growth, and irrigation anomalies. However, this manual interpretation method still relies on the operator's experience and cannot achieve high-frequency, standardized automated analysis of large-scale farmland. Furthermore, some existing semi-automated farmland monitoring methods often separate image processing from agronomic decision-making, resulting in intermediate outputs that lack correlation with specific water and fertilizer management units, making it difficult for monitoring information to directly guide field operations.

[0003] Existing technologies have several shortcomings in key aspects of farmland management. Regarding the construction of basic spatial data for farmland, manual land surveying is inefficient and subject to subjective boundary determination, lacking methods for automatically generating and dynamically updating standardized spatial base maps of farmland. In terms of irrigation facility inspection, leaks and blockages in underground or surface drip irrigation pipes typically rely on personnel walking along the pipes, which is time-consuming and results in delayed problem detection. Existing UAV remote sensing technology has not yet developed an automated identification process for irrigation anomalies. Regarding seedling management, the detection of missing seedlings, broken rows, and weed distribution is mostly based on manual sampling surveys, lacking high-resolution individual identification and spatial positioning methods that can cover the entire field.

[0004] In determining growth stages and providing early warning of diseases, existing methods often rely solely on meteorological data or single vegetation indices, failing to effectively integrate actual crop growth rhythms with disease occurrence dynamics models. Regarding nutrient management, traditional soil sampling and testing methods suffer from poor timeliness and low spatial resolution, while single remote sensing indices are insufficient to exclude interference from factors such as water stress. For lodging damage assessment, post-disaster visual observation or manual measurement is not only dangerous but also inefficient, lacking automated quantitative assessment methods based on three-dimensional pre- and post-disaster information. Finally, the outputs of existing monitoring modules are often scattered charts or raster images, lacking a systematic report generation mechanism capable of integrating multi-source anomaly information and generating executable operational instructions based on agronomic priorities. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an integrated air-ground agricultural monitoring method based on intelligent field patrols. It aims to automatically generate a standardized farmland spatial base map from raw UAV imagery through a unified technical process. Based on this base map, it integrates multiple functional modules, including irrigation anomaly identification, seedling and weed detection, growth stage determination and disease risk assessment, nutrient status monitoring, lodging detection, agricultural condition suggestion generation, and patrol route planning. This method does not rely on multi-source data fusion or complex model training processes. Instead, it transforms remote sensing observations into agricultural condition diagnostic conclusions and action instructions tightly coupled with farmland management units through a series of explicit spatial analysis, image processing, and rule-based reasoning steps. By establishing a complete chain from data acquisition to decision output, this invention can provide large-scale planting entities with interpretable, traceable, and directly applicable technical solutions for guiding field operations, thereby improving the automation and precision of farmland management.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] A method for integrated air-ground agricultural monitoring based on intelligent field patrol includes the following steps:

[0008] S1, acquire a sequence of UAV images of the target farmland, the sequence of images including visible light images and thermal infrared images;

[0009] S2, Spatial registration and geometric correction are performed on the UAV image sequence to generate orthophotos and digital surface models covering the target farmland;

[0010] S3, Based on the texture features of the orthophoto and the elevation change information of the digital surface model, extract the planting boundary of the farmland and identify the field ridges, roads and ditches inside the farmland;

[0011] S4. Based on the planting boundary and the internal structure, and combined with the connected component analysis algorithm, the farmland area is divided into multiple water and fertilizer management units, each of which corresponds to an independent spatial vector polygon.

[0012] S5. Based on the spatial vector polygon of the water and fertilizer management unit, a standardized farmland spatial base map is generated. The base map includes the boundary, area and spatial topology of each unit.

[0013] Further, S4 includes:

[0014] Based on the pixel connectivity within the planting boundary, continuous crop planting areas are identified as initial units;

[0015] For initial cells with an area lower than a preset threshold, an adjacent cell merging operation is performed, merging them into neighboring cells that share the longest common boundary;

[0016] For initial units with an area greater than a preset threshold, they are divided based on vector lines of internal field ridges or ditches to generate multiple sub-units;

[0017] Assign a unique identifier to each of the final generated water and fertilizer management units and record its geometric center coordinates and boundary length.

[0018] Furthermore, after generating the orthophoto and digital surface model covering the target farmland in step S2, the process also includes an irrigation anomaly identification step:

[0019] Obtain the canopy temperature distribution map corresponding to the thermal infrared image, and obtain the air temperature recorded by the synchronous meteorological station;

[0020] Calculate the canopy temperature difference value for each pixel in the canopy temperature distribution map;

[0021] Spatial continuity analysis was performed on the crown temperature difference values ​​to extract abnormal patches that form continuous closed regions;

[0022] Based on the spatial morphological characteristics of the abnormal patches, abnormal patches that extend linearly along the irrigation pipeline are identified as suspected leakage areas, and abnormal patches that are isolated points or small clusters with a crown temperature difference higher than the surrounding environment are identified as suspected blockage areas.

[0023] Output a risk distribution map that includes the geographical location and spatial extent of the suspected leak area and the suspected blockage area.

[0024] Furthermore, after extracting the planting boundary of the farmland based on the texture features of the orthophoto and the elevation change information of the digital surface model in step S3, the process also includes a seedling anomaly identification step:

[0025] Acquire high-resolution visible light orthophotos of crop seedlings;

[0026] An instance segmentation algorithm is performed on the high-resolution visible light orthophoto to identify individual crop pixel regions and background pixel regions in the image, wherein the background pixel regions include soil pixels and weed pixels.

[0027] Based on the individual pixel regions of the crops, the geometric center point of each crop is extracted, and spatial clustering is performed on all center points to fit and generate the center line of the crop row.

[0028] Along the center line of the crop row, the Euclidean distance between the center points of two adjacent crop individuals is calculated as the measured plant spacing, and the measured plant spacing is compared with the preset reference plant spacing. The positions where the measured plant spacing is greater than the upper limit of the reference plant spacing are marked as missing seedlings.

[0029] Texture feature analysis is performed on the background pixel area. Areas with texture complexity lower than the preset weed threshold are identified as bare soil areas, and areas with texture complexity higher than or equal to the preset weed threshold are identified as weed areas. A weed risk distribution map is then generated.

[0030] Furthermore, after generating the standardized farmland spatial base map in step S5, the process also includes a growth period and disease risk assessment step:

[0031] Acquire multi-temporal orthophotos of the target farmland at multiple consecutive time points, and calculate the wide dynamic range vegetation index map for each time point;

[0032] Based on the wide dynamic vegetation index map, a time-series curve of vegetation index change over time is constructed within each water and fertilizer management unit.

[0033] Extract the monotonically increasing intervals, peak points, and monotonically decreasing intervals from the time-series curve, and map the curve interval where the current time point is located to the seedling stage, jointing stage, heading stage, or maturity stage according to the preset phenological stage criteria.

[0034] Acquire daily meteorological data of the target farmland, including daily maximum temperature, daily minimum temperature and relative humidity, and determine the host sensitivity coefficient based on the currently determined growth stage;

[0035] Based on the daily meteorological data, the host sensitivity coefficient, and the preset disease occurrence conditions, the disease infection index of each water and fertilizer management unit is calculated, and the disease infection index is divided into low-risk, medium-risk, and high-risk levels.

[0036] Furthermore, after generating the standardized farmland spatial base map in step S5, a nutrient monitoring step is also included:

[0037] Acquire multi-band orthophotos of the target farmland, wherein the multi-bands include at least the near-infrared band and the red-edge band;

[0038] Based on the reflectance values ​​of the near-infrared band and the red-edge band, the chlorophyll index value of each pixel is calculated to generate a nutrient status distribution map.

[0039] Local spatial autocorrelation analysis was performed on the nutrient status distribution map to identify low-value clusters with nutrient index values ​​significantly lower than the neighborhood mean and high-value clusters with nutrient index values ​​significantly higher than the neighborhood mean.

[0040] The low-value clustering areas are marked as potential nutrient deficiency areas, and the high-value clustering areas are marked as potential over-fertilization or fertilizer damage areas.

[0041] The potential nutrient deficiency areas, potential over-fertilization or fertilizer damage areas are overlaid with the spatial vector polygons of the water and fertilizer management units to generate an independent list of nutrient anomalies for each unit.

[0042] Furthermore, after generating the orthophoto and digital surface model covering the target farmland in step S2, a lodging detection step is also included:

[0043] Acquire the first orthophoto and the first digital surface model in the pre-disaster phase, and the second orthophoto and the second digital surface model in the post-disaster phase;

[0044] The first digital surface model and the second digital surface model are subjected to pixel-by-pixel difference calculation to generate a canopy height change map, wherein the area with a negative pixel value indicates a decrease in canopy height;

[0045] Extract all consecutive pixel regions whose height decrease exceeds a preset threshold from the canopy height change map as candidate lodging regions;

[0046] For each candidate collapsed area, calculate its first texture directionality measure on the first orthophoto and its second texture directionality measure on the second orthophoto.

[0047] The region where the second texture directionality metric is significantly attenuated compared to the first texture directionality metric, and where the spatial overlap rate between the significantly attenuated region and the candidate collapsed region exceeds a preset overlap threshold, is ultimately determined to be a confirmed collapsed region.

[0048] The total number of pixels in the confirmed lodging area is counted, and the lodging area is calculated by combining the spatial resolution of the second orthophoto. Based on the average height reduction value of the canopy height change map in the confirmed lodging area, the lodging is classified into mild, moderate, and severe levels.

[0049] Furthermore, following step S5, a step for generating agricultural information suggestions is also included:

[0050] Receives the irrigation anomaly area layer output by the irrigation anomaly identification module, the seedling missing point and weed area layer output by the seedling anomaly identification module, the disease risk zoning layer output by the disease risk assessment module, the nutrient anomaly zoning layer output by the nutrient monitoring module, and the lodging area layer output by the lodging detection module.

[0051] All received layers are spatially overlaid with the water and fertilizer management units in the standardized farmland spatial base map to determine the types of abnormal events contained in each water and fertilizer management unit and their spatial coverage.

[0052] For a water and fertilizer management unit that includes multiple types of abnormal events, the primary management problem of the unit is determined based on a preset agronomic treatment priority rule base. The priority rule base defines the sorting logic of high disease risk taking precedence over nutrient abnormality, nutrient abnormality taking precedence over irrigation abnormality, and irrigation abnormality taking precedence over weed risk.

[0053] For each water and fertilizer management unit, based on its primary management problem, the corresponding handling operation text is retrieved from the standardized operation instruction library. The handling operation text includes a problem location description and action instructions.

[0054] Furthermore, the method also includes a visualization report generation step:

[0055] Create a blank geographic information system layout;

[0056] In the first area of ​​the layout, the standardized farmland spatial base map is loaded and displayed, and the primary management problem types of each water and fertilizer management unit are overlaid in a layered color scheme.

[0057] In the second area of ​​the layout, a list of agricultural tasks is generated in the form of a list. Each item in the list corresponds to a water and fertilizer management unit and includes the unit number, the type of primary management problem, and the corresponding handling operation text.

[0058] In the third area of ​​the layout, for the selected item in the agricultural task list, the original orthophoto local slice of the water and fertilizer management unit to which the unit belongs is dynamically loaded and displayed, and the specific spatial location of each abnormal event in the unit is marked with a highlighted boundary on the slice.

[0059] Output the agricultural monitoring report in a portable document format.

[0060] According to a second aspect of the present invention, the present invention claims protection for an integrated air-ground agricultural monitoring system based on intelligent field patrol, comprising:

[0061] One or more processors;

[0062] A memory that stores one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the aforementioned integrated air-ground agricultural monitoring method based on intelligent field patrol.

[0063] This invention relates to an integrated air-ground agricultural monitoring method and system based on intelligent field patrol, belonging to the fields of agricultural informatization and UAV remote sensing monitoring technology. The method involves acquiring a sequence of UAV imagery of the target farmland; performing spatial registration and geometric correction on the UAV imagery sequence to generate an orthophoto and digital surface model covering the target farmland; extracting the planting boundaries of the farmland based on elevation change information and identifying the field ridges, roads, and ditches within the farmland; dividing the farmland area into multiple water and fertilizer management units, corresponding to independent spatial vector polygons, using a connected component analysis algorithm; and generating a standardized farmland spatial base map based on the spatial vector polygons, including the boundaries, area, and spatial topological relationships of each unit. This invention automates the entire process from image acquisition to agricultural condition decision-making, effectively reducing the intensity of manual field patrols, improving the timeliness and spatial accuracy of farmland anomaly identification, and providing directly executable operational instructions for large-scale precision management. Attached Figure Description

[0064] Figure 1 A flowchart illustrating the workflow of an integrated air-ground agricultural monitoring method based on intelligent field patrol, as claimed in an embodiment of the present invention.

[0065] Figure 2 The second working flowchart of a sky-ground integrated agricultural monitoring method based on intelligent field patrol, as claimed in an embodiment of the present invention, is shown below.

[0066] Figure 3 The third workflow diagram is for a sky-ground integrated agricultural monitoring method based on intelligent field patrol, which is claimed in an embodiment of the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0068] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of those features. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications in the embodiments of this application, such as up, down, left, right, front, back, etc., are only used to explain the relative positional relationships and movements between components in a specific orientation as shown in the accompanying drawings. If the specific orientation changes, the directional indications will change accordingly. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0069] References to embodiments herein mean that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0070] According to a first embodiment of the present invention, the present invention claims protection for a sky-ground integrated agricultural condition monitoring method based on intelligent field patrol, referring to... Figure 1 This includes the following steps:

[0071] S1, acquire a sequence of UAV images of the target farmland, the sequence of images including visible light images and thermal infrared images;

[0072] S2, Spatial registration and geometric correction are performed on the UAV image sequence to generate orthophotos and digital surface models covering the target farmland;

[0073] S3, Based on the texture features of the orthophoto and the elevation change information of the digital surface model, extract the planting boundary of the farmland and identify the field ridges, roads and ditches inside the farmland;

[0074] S4. Based on the planting boundary and the internal structure, and combined with the connected component analysis algorithm, the farmland area is divided into multiple water and fertilizer management units, each of which corresponds to an independent spatial vector polygon.

[0075] S5. Based on the spatial vector polygon of the water and fertilizer management unit, a standardized farmland spatial base map is generated. The base map includes the boundary, area and spatial topology of each unit.

[0076] In this embodiment, the operator first uses a multi-rotor UAV equipped with a visible light camera and a thermal infrared camera to perform a planned flight mission over the target farmland. The UAV automatically flies according to preset overlap rates, such as a forward overlap rate of 80%, a lateral overlap rate of 70%, and a fixed flight altitude, such as 100 meters above the ground, continuously acquiring ground images. During the flight, the UAV's onboard global navigation satellite system receiver records the precise three-dimensional spatial coordinates of each image capture moment in real time, while the inertial measurement unit simultaneously records the UAV's attitude angle data, including roll angle, pitch angle, and heading angle. After the flight mission is completed, all visible light image files, thermal infrared image files, and corresponding positioning and attitude data files stored on the UAV's memory card are imported into the ground processing workstation.

[0077] After receiving the aforementioned images and positioning / attitude data, the ground processing workstation first performs lens distortion correction on each image to eliminate image edge distortion caused by the optical characteristics of the camera lens. Then, based on the image's shooting position and attitude angle, it uses a motion reconstruction structure algorithm to perform joint adjustment calculations on all images, accurately estimating the true exterior orientation elements of each image relative to the ground. Next, using the calculated image parameters, it extracts corresponding feature points between images and generates 3D point cloud data through multi-view dense matching. Each point in the point cloud data contains three coordinate information: longitude, latitude, and elevation. The 3D point cloud data is projected onto a horizontal reference plane, and the data-free areas are filled using an interpolation algorithm to generate a digital surface model. This model records the top elevation information of all objects on the ground, including crops, buildings, and the ground surface. At the same time, each visible light image and thermal infrared image is resampled and stretched according to its exterior orientation elements to eliminate geometric distortions caused by terrain undulations and camera tilt. All the transformed images are then stitched and fused according to their actual ground positions to form a seamless orthophoto covering the entire target farmland. Each pixel in this orthophoto carries real geographic coordinates and has a vertical downward viewing angle effect.

[0078] The workstation reads the generated visible light orthophoto image, converts it to grayscale, and calculates the degree of grayscale change for each pixel and its neighboring pixels to obtain a texture feature map. Simultaneously, it reads the digital surface model and calculates the gradient of the elevation value at each pixel relative to the elevation values ​​of its eight surrounding neighboring pixels, generating an elevation gradient map. The texture feature map and the elevation gradient map are then weighted and superimposed to generate a comprehensive edge response map. Non-maximum suppression is performed on the comprehensive edge response map to refine the edge lines. Then, a hysteresis thresholding method is used to identify continuous and closed edge curves as ground feature boundaries. Within these boundaries, based on the area of ​​the closed boundary region and the consistency of the texture within the region, areas with larger farmland planting boundaries and relatively uniform internal texture are distinguished from areas with narrow or linear areas such as field ridges, roads, and ditches with significant elevation abrupt changes. The identified planting boundaries are converted into vector polygons and stored in memory.

[0079] The workstation uses the extracted planting boundary polygon as a mask, treating the farmland area inside the boundary as an image to be segmented. First, it identifies the field ridges, roads, and ditches obtained in step 3 within this area, using these vector lines as natural segmentation barriers to divide the area into multiple unconnected connected domains, each of which serves as an initial management unit. Then, it iterates through all initial management units, calculating the area of ​​each unit, which is the number of pixels it contains multiplied by the ground area represented by each pixel. For units with an area smaller than a preset threshold (e.g., the minimum operating unit area), the workstation finds the neighboring unit with the longest shared boundary length, merges all pixels of the smaller unit into the neighboring unit, and updates the boundary vector of the neighboring unit. For units with an area larger than a preset maximum threshold, the workstation checks whether there are artificial facilities inside the unit that can be used as a segmentation basis, such as the center line of the main irrigation pipe. If so, it generates a segmentation line along both sides of the center line of the facility at a fixed buffer distance (e.g., 1 meter), dividing the unit into two sub-units. After segmentation, a unique identifier is generated for each water and fertilizer management unit, the latitude and longitude coordinates of its geometric center point are calculated, the perimeter and area of ​​its boundary closed polygon are calculated, and these attribute information are associated with the corresponding vector polygon.

[0080] The workstation organizes the vector polygons and their attribute data of all water and fertilizer management units into a feature class. Simultaneously, it records the adjacency and containment relationships between the polygons; for example, polygon A's north side is adjacent to polygon B, or polygon C is completely surrounded by polygon D. This feature class is output in a common Geographic Information System (GIS) vector format such as Shapefile or GeoJSON, along with a corresponding metadata file containing map projection parameters, coordinate reference system information, data production time, and version number. Finally, this vector file and metadata file together constitute a standardized farmland spatial base map, which can be directly loaded and used by downstream analysis modules for spatial queries, overlay analysis, and regional statistics.

[0081] Further, S4 includes:

[0082] Based on the pixel connectivity within the planting boundary, continuous crop planting areas are identified as initial units;

[0083] For initial cells with an area lower than a preset threshold, an adjacent cell merging operation is performed, merging them into neighboring cells that share the longest common boundary;

[0084] For initial units with an area greater than a preset threshold, they are divided based on vector lines of internal field ridges or ditches to generate multiple sub-units;

[0085] Assign a unique identifier to each of the final generated water and fertilizer management units and record its geometric center coordinates and boundary length.

[0086] In this embodiment, the workstation acquires the planting boundary vector polygon extracted in step 3 and uses the region inside the polygon as an analysis mask. It reads the visible light orthophoto image within the mask area and categorizes all pixels in the image into crop pixels and non-crop pixels. The classification criteria can be based on a trained classifier, but only the specific judgment logic is described here: if a pixel's green channel value is significantly higher than its red and blue channel values, and its normalized difference vegetation index is higher than an empirical threshold, it is classified as a crop pixel; otherwise, it is a non-crop pixel. Then, the workstation uses a four-connectivity or eight-connectivity rule to traverse all crop pixels: starting from the first crop pixel in the upper left corner, it is marked as the current connected region, and its adjacent crop pixels (up, down, left, right, or diagonally) are recursively added to the same connected region until no new adjacent crop pixels can be added. When a connected region can no longer be expanded, the set of pixels contained in that region is recorded, and the search continues for the next unmarked crop pixel, repeating the above process. Finally, each continuous plot area composed of connected crop pixels is defined as an initial water and fertilizer management unit.

[0087] The workstation presets a minimum manageable unit area threshold, which is set by the user based on the actual operating width of agricultural machinery, such as the working width of a sprayer or harvester. It iterates through all initial water and fertilizer management units, calculates the pixel area of ​​each unit, and compares it to this threshold. When the area of ​​a unit is below the threshold, the workstation performs a merging decision: First, it calculates the common boundary length between the small unit and all its neighboring units (units sharing at least one common boundary pixel). Then, it identifies the neighboring unit with the longest common boundary length. Finally, it modifies the labels of all pixels in the small unit to the labels of the selected neighboring units, merges the boundary point sequence of the small unit into the boundary point sequence of the neighboring units, and updates the area statistics of the neighboring units. If a small unit has no neighboring units and is an isolated patch, it is retained as is and recorded as a special unit. This process is repeated until all units with areas below the threshold have been checked and processed.

[0088] The workstation presets a maximum manageable unit area threshold, which is set by the user based on the application accuracy range of the irrigation valve group or fertilizer prescription map. It iterates through all merged units and calculates the area of ​​each unit. When the area of ​​a unit exceeds this threshold, the workstation initiates a segmentation process: First, it extracts elevation data within the unit's area from the original digital surface model generated in step 1 and performs slope analysis to generate a slope map. Second, it identifies whether there are existing field ridges or ditch vector lines within the unit. If so, it generates a segmentation boundary along the centerline of these existing linear features, dividing the unit in two. If no obvious linear features exist, the workstation calculates the geometric center point of the unit and then uses a straight line passing through this center point and perpendicular to the crop planting row direction as the initial segmentation line to divide the unit into two sub-units with as equal areas as possible. After segmentation, a new vector polygon is generated for each sub-unit, inheriting the original unit's identifier prefix and adding a sub-number.

[0089] After the above merging and segmentation operations, the workstation obtains the final set of cells. For each cell in the set, the workstation assigns a globally unique identifier according to its spatial order from top to bottom and from left to right. This identifier consists of a farmland code, the letter U, and three digits, for example, Field01-U001. Then, for each cell, the workstation performs the following attribute calculations: traverse the sequence of vertices constituting the cell boundary and calculate the average longitude and average latitude of all vertices as the geometric center coordinates; traverse all adjacent boundary vertices and accumulate the Euclidean distance between every two adjacent vertices to obtain the total boundary length; count the number of pixels contained within the cell and multiply it by the ground area corresponding to a single pixel in the orthophoto to obtain the actual area in square meters; finally, these attribute values ​​are stored in a memory table as key-value pairs associated with the cell identifier.

[0090] Furthermore, after generating the orthophoto and digital surface model covering the target farmland in step S2, the process also includes an irrigation anomaly identification step:

[0091] Obtain the canopy temperature distribution map corresponding to the thermal infrared image, and obtain the air temperature recorded by the synchronous meteorological station;

[0092] Calculate the canopy temperature difference value for each pixel in the canopy temperature distribution map;

[0093] Spatial continuity analysis was performed on the crown temperature difference values ​​to extract abnormal patches that form continuous closed regions;

[0094] Based on the spatial morphological characteristics of the abnormal patches, abnormal patches that extend linearly along the irrigation pipeline are identified as suspected leakage areas, and abnormal patches that are isolated points or small clusters with a crown temperature difference higher than the surrounding environment are identified as suspected blockage areas.

[0095] Output a risk distribution map that includes the geographical location and spatial extent of the suspected leak area and the suspected blockage area.

[0096] In this embodiment, the workstation loads the thermal infrared orthophoto image generated in step 2. Each pixel value in the image represents the radiance value of the ground object in the thermal infrared band. The workstation uses a pre-calibrated radiance calibration coefficient to convert the radiance value of each pixel into a Celsius temperature value, thereby generating a canopy temperature distribution map covering the entire target farmland. At the same time, the workstation reads the data file recorded by the weather station deployed in the center of the farmland through the serial communication interface, extracts the air temperature measurement values ​​within ten minutes before and after the UAV flight time, and calculates the arithmetic mean of these measurements as the air temperature representing the flight.

[0097] The workstation creates a new image of the same size as the canopy temperature distribution map as the canopy-air temperature difference map. It iterates through each pixel in the canopy temperature distribution map, reads the canopy temperature value of that pixel, and then subtracts the canopy temperature value from the air temperature value or subtracts the air temperature from the canopy temperature value. The specific calculation result is stored in the corresponding position in the canopy-air temperature difference map according to the sign convention of the subsequent judgment logic. In this difference map, positive value areas indicate that the canopy temperature is higher than the air temperature, and negative value areas indicate that the canopy temperature is lower than the air temperature.

[0098] The workstation performs binarization processing on the crown temperature difference map: An anomaly threshold range is set, for example, the difference is greater than a specific positive value or less than a specific negative value. Pixels within this range are marked as candidate anomaly pixels and assigned a value of 1, while the remaining pixels are marked as normal pixels and assigned a value of 0. Then, a connected component analysis algorithm is used to scan the binary image, merging all pixels with a value of 1 and that are interconnected into an independent anomaly patch. For each detected patch, the workstation further checks whether its shape forms a closed region—that is, whether the boundary pixels of the patch are connected end to end to form a loop without gaps. For unclosed patches, the workstation performs morphological closing operations, first dilating and then eroding, to attempt to fill the gaps and close them. Finally, each continuously closed anomaly patch is recorded as an anomaly object, storing the coordinate sequence of its boundary pixels.

[0099] The workstation acquires a vector line layer of all irrigation pipes within the target farmland. This layer needs to be imported into the system beforehand and includes the location and direction of each pipe. For each abnormal patch extracted in step 3, the workstation calculates its minimum bounding rectangle and obtains the direction of the long side and the aspect ratio. Then, it calculates the distance from the geometric center of the patch to the nearest irrigation pipe. If this distance is less than a preset pipe association threshold, and the aspect ratio of the patch's minimum bounding rectangle is greater than a preset linear feature threshold (e.g., an aspect ratio greater than 5:1), and the angle between the long side direction and the pipe direction is less than a preset angle threshold (e.g., within 15 degrees), then the patch is identified as an abnormal patch extending linearly along the pipe direction and marked as a suspected leak area. If the aspect ratio of the patch's minimum bounding rectangle is less than the linear feature threshold, and the patch area is less than a preset small area threshold, and the crown temperature difference map inside the patch shows that the difference in the central area is significantly higher than that in the edge area (i.e., forming a heat island effect), then it is identified as an abnormal patch in the form of isolated points or small clusters and marked as a suspected blockage area. Patches that do not match any of the above features are marked as areas to be checked.

[0100] The workstation creates a new vector layer as output. For each patch marked as a suspected leak area, a polygon feature is created in the layer, and an attribute field "Anomaly Type" is added with the value "Leakage," and a priority field is added with the value "High." For each patch marked as a suspected blockage area, a polygon feature is also created, with the attribute "Blockage Type" set to "Blockage" and the priority set to "Medium." For areas awaiting verification, the priority is set to "Low." The workstation saves all polygon features along with their geographic coordinates into a Shapefile file. Simultaneously, a text-formatted location list is generated, sequentially listing the center point coordinates, area size, and criteria for each anomalous patch, providing navigation aids for field personnel to the site for verification.

[0101] Furthermore, referring to Figure 2In step S3, after extracting the planting boundary of farmland based on the texture features of the orthophoto and the elevation change information of the digital surface model, a seedling anomaly identification step is also included:

[0102] Acquire high-resolution visible light orthophotos of crop seedlings;

[0103] An instance segmentation algorithm is performed on the high-resolution visible light orthophoto to identify individual crop pixel regions and background pixel regions in the image, wherein the background pixel regions include soil pixels and weed pixels.

[0104] Based on the individual pixel regions of the crops, the geometric center point of each crop is extracted, and spatial clustering is performed on all center points to fit and generate the center line of the crop row.

[0105] Along the center line of the crop row, the Euclidean distance between the center points of two adjacent crop individuals is calculated as the measured plant spacing, and the measured plant spacing is compared with the preset reference plant spacing. The positions where the measured plant spacing is greater than the upper limit of the reference plant spacing are marked as missing seedlings.

[0106] Texture feature analysis is performed on the background pixel area. Areas with texture complexity lower than the preset weed threshold are identified as bare soil areas, and areas with texture complexity higher than or equal to the preset weed threshold are identified as weed areas. A weed risk distribution map is then generated.

[0107] In this embodiment, the workstation acquires high-resolution visible light orthophotos collected by the drone during the period from crop emergence to jointing; the ground resolution of the image is better than 2 centimeters per pixel to ensure that the shape of individual crops can be clearly distinguished; the image has undergone geometric correction and stitching processing in step 2 and has accurate geographic coordinates.

[0108] The workstation loads the high-resolution orthophoto and divides it into fixed-size image patches, such as 256 pixels × 256 pixels. For each image patch, the workstation performs forward computation on a pre-trained instance segmentation neural network model, which outputs the probability that each pixel in the image patch belongs to a crop individual and the probability that it belongs to the background. For each pixel, if its probability of belonging to a crop individual is higher than its probability of belonging to the background, the pixel is marked as a crop pixel; otherwise, it is marked as a background pixel. The set of all crop pixels forms the crop individual pixel region, and the set of all background pixels forms the background pixel region. Further, within the background pixel region, the workstation uses another classifier to further subdivide the background pixels into soil and weeds, based on the position of the combined vector of the pixel's color and texture features in the feature space. Finally, three binary masks are output: a crop mask, a soil mask, and a weed mask.

[0109] The workstation acquires a crop mask image. First, morphological erosion removes isolated, excessively small clusters of crop pixels, typically caused by noise or scattered weeds. Then, for the remaining clusters, the centroid pixel of each cluster is extracted as the geometric center point of the individual crop. Using all center points as input points, the workstation employs the Hough transform algorithm to detect straight line patterns within these points. Specifically, each point is mapped to the polar radius-polar angle parameter space, and an accumulator counts the parameters corresponding to local maxima; these parameters represent the fitted crop row lines. For each fitted line, the workstation expands it into a centerline of a certain width, the direction of which is the crop row direction, with the centerline positioned at the center of the crop row.

[0110] The workstation slides a fixed-length window along the centerline of each crop row at a preset step size equal to half the theoretical plant spacing. At each window location, the number of crop individual center points contained within the window is counted. If this number is less than the preset expected number (the expected number equals the window length divided by the theoretical plant spacing), a missing seedling is determined at that window location. Then, the workstation calculates the measured distance between the center point of the window and the two adjacent confirmed crop individual center points. If this measured distance is greater than the upper limit of the theoretical plant spacing, such as 1.5 times the theoretical plant spacing, the location of the window center point is marked as a missing seedling point. All missing seedling points are collected to generate a point feature layer, with each point accompanied by a field recording the length range of the missing seedling.

[0111] The workstation acquires both a weed mask and a soil mask. For each continuous connected region of weeds within the weed mask, the workstation calculates the region's area and boundary complexity (boundary perimeter divided by the square root of the area). Simultaneously, at the corresponding soil mask location, it calculates the local binary pattern feature value of the soil texture. The workstation uses the area, boundary complexity, and surrounding soil texture feature value of the weed region as input to look up a predefined mapping table, which outputs a risk level of low, medium, or high. For example, a large area with complex boundaries and uniform surrounding soil texture is marked as high-risk. Then, the workstation divides the target farmland into multiple small blocks using a grid, such as a 10m x 10m grid. For each block, it calculates the percentage of high-risk weed regions within it. If the percentage exceeds a preset threshold, the block is marked as a high-risk weed region. Finally, a raster layer is generated, where the value of each pixel represents the weed risk level at that location.

[0112] Furthermore, after generating the standardized farmland spatial base map in step S5, the process also includes a growth period and disease risk assessment step:

[0113] Acquire multi-temporal orthophotos of the target farmland at multiple consecutive time points, and calculate the wide dynamic range vegetation index map for each time point;

[0114] Based on the wide dynamic vegetation index map, a time-series curve of vegetation index change over time is constructed within each water and fertilizer management unit.

[0115] Extract the monotonically increasing intervals, peak points, and monotonically decreasing intervals from the time-series curve, and map the curve interval where the current time point is located to the seedling stage, jointing stage, heading stage, or maturity stage according to the preset phenological stage criteria.

[0116] Acquire daily meteorological data of the target farmland, including daily maximum temperature, daily minimum temperature and relative humidity, and determine the host sensitivity coefficient based on the currently determined growth stage;

[0117] Based on the daily meteorological data, the host sensitivity coefficient, and the preset disease occurrence conditions, the disease infection index of each water and fertilizer management unit is calculated, and the disease infection index is divided into low-risk, medium-risk, and high-risk levels.

[0118] In this embodiment, the workstation retrieves all visible light and multispectral orthophoto sequences from a database or file system within the past two months, at uniform time intervals, such as every 5 days. For each time point in the image, the workstation reads its near-infrared and red band reflectance data. For each pixel in the image, a wide dynamic range vegetation index (WDR) value is calculated, which is obtained by dividing the difference between the near-infrared and red reflectance by their sum and then multiplying by a scaling factor. After calculation, a WDR map corresponding to that time point is generated. The size of this map is consistent with the original orthophoto image, and the pixel values ​​are floating-point numbers.

[0119] The workstation loads the standardized farmland spatial base map generated in step 5 to obtain the vector boundary of each water and fertilizer management unit. For each unit, the workstation traverses the wide dynamic vegetation index map at all time points. At each time point, the wide dynamic vegetation index value of all pixels inside the boundary of the unit is extracted, and the arithmetic mean of these values ​​is calculated as the representative vegetation index of the unit at that time point. The representative vegetation indices of all time points are arranged in chronological order to form a numerical sequence, and the time point corresponding to each value is marked on the horizontal axis, i.e., the specific date. The sequence is stored in the form of a line graph, with the horizontal axis representing time and the vertical axis representing the vegetation index value, thus forming the time series curve of the unit.

[0120] The workstation smooths the time-series curves of each unit, eliminating local fluctuations caused by cloud cover or sensor noise. On the smoothed curves, the workstation uses the first-order difference method to find that the first-order difference is positive in the rising segment and negative in the falling segment, and locates the point where the first-order difference turns from positive to negative as the peak point. Then, the workstation sets a baseline threshold, such as 10% of the maximum value of the entire cycle. Starting from the beginning of the curve, when the vegetation index value first exceeds this threshold, it is marked as the beginning of the seedling stage. Starting from the peak point, it searches to the right. When the vegetation index value first drops to a certain proportion of the peak, such as 80%, it is marked as the end of the heading stage and the beginning of the maturity stage. When the vegetation index value drops to close to the baseline threshold, it is marked as the end of the maturity stage. Based on the position of the current date on the curve, the workstation determines which interval the current date falls into, and thus outputs the current growth stage.

[0121] The workstation accesses the meteorological data service platform via a network interface. Based on the latitude and longitude coordinates of the target farmland, it downloads daily meteorological data records from the sowing date to the current date. Each record includes the daily maximum temperature, daily minimum temperature, and daily average relative humidity. Simultaneously, the workstation reads the currently determined growth stage. Internally, the workstation maintains a lookup table listing the host sensitivity coefficients for different crop varieties at various growth stages. For example, for corn, the sensitivity coefficient is highest during the tasseling stage and lower during the emergence and maturity stages. The workstation looks up the corresponding value in the lookup table based on the current growth stage. If the current growth stage is tasseling, a higher sensitivity coefficient is set; otherwise, a lower value is set.

[0122] The workstation loads a rule base for disease occurrence conditions specific to the target crop species. For calculating the disease infection index, the workstation employs the following iterative logic: An initial disease level is set to a small value, such as 0.001, and then iterates daily. Each day, the average temperature is calculated based on the day's highest and lowest temperatures. Then, a temperature modulation function table is consulted to obtain a temperature influence factor between 0 and 1. The relative humidity is consulted to obtain a humidity influence factor. The duration of leaf wetness is calculated based on the day's rainfall, and then a wetness duration modulation function table is consulted to obtain a wetness influence factor. The sensitivity coefficient, temperature influence factor, humidity influence factor, and wetness influence factor are multiplied together, and then multiplied by the inherent disease propagation coefficient to obtain the day's propagation rate. Finally, the current disease level is multiplied by 1, subtracted from the current disease level, and then multiplied by the propagation rate to obtain the day's increment in disease level. The increment is added to the current disease level to obtain a new disease level; the above process is repeated until the current date, and the final disease level is the disease infection index; the workstation compares this index with a preset threshold range: below the first threshold, for example, 0.3, is low risk; between the first and second thresholds, for example, 0.3 to 0.7, is medium risk; above the second threshold, is high risk; finally, the workstation generates a risk level label for each water and fertilizer management unit and outputs a risk zoning map, in which different risk levels are represented by different colors.

[0123] Furthermore, after generating the standardized farmland spatial base map in step S5, a nutrient monitoring step is also included:

[0124] Acquire multi-band orthophotos of the target farmland, wherein the multi-bands include at least the near-infrared band and the red-edge band;

[0125] Based on the reflectance values ​​of the near-infrared band and the red-edge band, the chlorophyll index value of each pixel is calculated to generate a nutrient status distribution map.

[0126] Local spatial autocorrelation analysis was performed on the nutrient status distribution map to identify low-value clusters with nutrient index values ​​significantly lower than the neighborhood mean and high-value clusters with nutrient index values ​​significantly higher than the neighborhood mean.

[0127] The low-value clustering areas are marked as potential nutrient deficiency areas, and the high-value clustering areas are marked as potential over-fertilization or fertilizer damage areas.

[0128] The potential nutrient deficiency areas, potential over-fertilization or fertilizer damage areas are overlaid with the spatial vector polygons of the water and fertilizer management units to generate an independent list of nutrient anomalies for each unit.

[0129] In this embodiment, the workstation loads the generated multi-band orthophoto image, which includes at least the near-infrared band, the red-edge band with a wavelength of approximately 700-730 nanometers, and the green band; the image has been radiometrically calibrated, and the pixel values ​​represent the surface reflectance.

[0130] The workstation creates a blank chlorophyll index map with the same dimensions as the image. It iterates through each pixel in the image, reading its reflectance values ​​in the near-infrared band and the red-edge band. Then, it subtracts the red-edge reflectance from the near-infrared reflectance and divides the difference by the red-edge reflectance to obtain the chlorophyll index value for that pixel. This value is written to the corresponding position on the chlorophyll index map. If the red-edge reflectance of a pixel is close to zero (e.g., below the noise threshold), the chlorophyll index value for that pixel is set to a special flag indicating invalidity. After calculation, the entire chlorophyll index map represents the spatial distribution of relative chlorophyll content in the crop canopy; a higher value generally indicates more abundant nitrogen nutrition.

[0131] The workstation performs raster data analysis on the chlorophyll index map. First, a movable window, such as a 15-pixel x 15-pixel square window, is defined. The center of the window is placed on each pixel of the chlorophyll index map, and the mean and standard deviation of the chlorophyll index values ​​for all pixels within the window are calculated. Then, the difference between the chlorophyll index value of the center pixel and the window mean is calculated, and this difference is divided by the window standard deviation to obtain a local autocorrelation statistic. If this statistic is positive and its absolute value exceeds a preset significance threshold, such as greater than 2, then the center pixel is marked. If a value is negative and its absolute value exceeds a threshold, it is marked as part of a low-value cluster. All connected pixel clusters marked as low-value clusters are counted, and each cluster represents a potential nutrient deficiency area. All connected pixel clusters marked as high-value clusters represent a potential over-fertilization or fertilizer damage area. It should be noted that areas near field edges or roads may produce false positives due to boundary effects. The workstation will compare these areas with the boundaries of the water and fertilizer management units to exclude clusters that cross the unit boundaries.

[0132] The workstation calculates the average chlorophyll index value for each detected low-value cluster and compares it with the average chlorophyll index value of the entire farmland. If the average value of the low-value cluster is lower than the threshold percentage of the overall field average (e.g., below 80%), it is identified as a nutrient-deficient area. For each high-value cluster, the average chlorophyll index value is calculated. If the percentage of the high-value cluster is higher than the threshold percentage of the overall field average (e.g., above 120%), it is identified as a potential over-fertilization area. If the high-value cluster also exhibits abnormally dark green leaves and curled leaf shapes, this can be determined by analyzing whether the ratio of green channel intensity to red channel intensity in the original RGB image of the area is abnormally high, and it is further identified as a fertilizer damage risk area.

[0133] The workstation loads the vector polygon of each water and fertilizer management unit from the standardized farmland spatial base map. For each unit, the workstation performs the following operations: First, it creates an empty list to store the abnormal records of that unit; then, it iterates through all the vector polygons of nutrient deficiency areas, determining whether the deficiency area intersects with the spatial range of the current unit; if they intersect, it calculates the proportion of the area of ​​the intersecting part to the total area of ​​the deficiency area; if the proportion exceeds a preset threshold, such as 30%, the deficiency area is considered to belong to the current unit, and its identification code and area are appended to the list; the same process is applied to areas of over-fertilization / fertilizer damage; finally, the workstation outputs each unit and its corresponding list of abnormal records as a JSON data structure, where each abnormal record contains the abnormality type (deficiency / over-fertilization / fertilizer damage), the boundary coordinates of the abnormal area, and the affected area.

[0134] Furthermore, referring to Figure 3 After generating the orthophoto and digital surface model covering the target farmland in step S2, the process also includes a lodging detection step:

[0135] Acquire the first orthophoto and the first digital surface model in the pre-disaster phase, and the second orthophoto and the second digital surface model in the post-disaster phase;

[0136] The first digital surface model and the second digital surface model are subjected to pixel-by-pixel difference calculation to generate a canopy height change map, wherein the area with a negative pixel value indicates a decrease in canopy height;

[0137] Extract all consecutive pixel regions whose height decrease exceeds a preset threshold from the canopy height change map as candidate lodging regions;

[0138] For each candidate collapsed area, calculate its first texture directionality measure on the first orthophoto and its second texture directionality measure on the second orthophoto.

[0139] The region where the second texture directionality metric is significantly attenuated compared to the first texture directionality metric, and where the spatial overlap rate between the significantly attenuated region and the candidate collapsed region exceeds a preset overlap threshold, is ultimately determined to be a confirmed collapsed region.

[0140] The total number of pixels in the confirmed lodging area is counted, and the lodging area is calculated by combining the spatial resolution of the second orthophoto. Based on the average height reduction value of the canopy height change map in the confirmed lodging area, the lodging is classified into mild, moderate, and severe levels.

[0141] In this embodiment, the workstation retrieves the last orthophoto and digital surface model of the target farmland acquired before a disaster event such as a typhoon, rainstorm, or hail from historical data storage; these are denoted as the first orthophoto and the first digital surface model, respectively. Simultaneously, it loads the first orthophoto and digital surface model acquired during the flight after the disaster event, denoted as the second orthophoto and the second digital surface model, respectively. Both sets of data have undergone geometric correction, have the same spatial resolution and the same coordinate reference system, ensuring pixel-level alignment.

[0142] The workstation creates a new raster layer with the same size as the first digital surface model to store canopy height variation values. It iterates through each pixel in the first digital surface model, reading the pre-disaster elevation value at that location, and the post-disaster elevation value at the same location in the second digital surface model. The workstation subtracts the pre-disaster elevation value from the post-disaster elevation value to obtain the difference. This difference is written to the corresponding pixel in the canopy height variation map. If the difference is negative, it indicates a decrease in the canopy height of the surface object at that location; if it is positive, it indicates an increase in height, possibly due to crop growth or noise. The workstation then counts and records the absolute values ​​of all negative areas in the canopy height variation map.

[0143] The workstation sets a height reduction threshold, for example, 0.3 meters, which corresponds to the expected canopy height reduction after normal crop lodging. It iterates through the canopy height change map and marks all pixels with a difference less than the negative threshold as pixels with significant height reduction. Then, a connected component analysis algorithm is used to identify connected regions in these marked pixels. For each connected region, the workstation calculates its area pixel count and shape compactness. It removes isolated noise regions with too small an area, such as less than 1 square meter, and regions with too narrow a shape that may correspond to field roads. All remaining connected regions are stored as candidate lodging regions, along with their boundary and internal pixel sets.

[0144] The workstation performs texture directionality analysis on each candidate collapsed area. First, it crops an image sub-block from the first orthophoto that completely overlaps with the boundary of the candidate collapsed area. Then, it performs a Fast Fourier Transform on this sub-block, converting the image from the spatial domain to the frequency domain to obtain a spectrogram. In the spectrogram, the energy concentration direction represents the main direction of texture in the image. The workstation calculates the angular distribution of energy in the spectrogram within the range of 0 to 180 degrees, finding the angle corresponding to the energy peak, which is used as the first texture directionality measure. Then, the same operation is performed on the sub-block at the same location in the second orthophoto to obtain the second texture directionality measure. In addition, the workstation also calculates the concentration of spectral energy at the peak angle for both sub-blocks, i.e., the ratio of peak energy to average energy, denoted as the first concentration and the second concentration, respectively.

[0145] The workstation compares the first and second texture directionality measures for each candidate lodging region. If the angle difference between the second and first directionality measures exceeds a preset turning threshold, such as greater than 30 degrees, it indicates a significant change in crop row direction, a typical characteristic of lodging. Simultaneously, it compares the second concentration with the first concentration. If the second concentration is less than the first concentration multiplied by an attenuation coefficient, such as 0.7, it indicates a significant reduction in texture regularity, also a sign of lodging. If both conditions are met simultaneously, or if at least one condition is met and the average height reduction in the canopy height variation map for that region exceeds 1.5 times the height reduction threshold, then the candidate lodging region is ultimately confirmed as a lodging region. The set of all confirmed lodging regions constitutes the final lodging distribution.

[0146] The workstation counts the total number of pixels in the confirmed lodged areas. Based on the ground area corresponding to a single pixel in the second orthophoto (e.g., each pixel represents 0.01 square meters), the number of pixels is multiplied by this area value to obtain the total lodged area in square meters. Then, for each individual confirmed lodged area, the workstation calculates the average value of the canopy height change map within that area, i.e., the average height reduction value. Threshold ranges are set for mild lodging (e.g., 0.3 meters to 0.5 meters), moderate lodging (e.g., 0.5 meters to 0.8 meters), and severe lodging (e.g., greater than 0.8 meters). Based on which range the average height reduction value of each area falls into, a corresponding lodging level is assigned. Finally, the workstation outputs a lodging assessment layer, where each lodged area polygon contains the fields for area and level.

[0147] Furthermore, following step S5, a step for generating agricultural information suggestions is also included:

[0148] Receives the irrigation anomaly area layer output by the irrigation anomaly identification module, the seedling missing point and weed area layer output by the seedling anomaly identification module, the disease risk zoning layer output by the disease risk assessment module, the nutrient anomaly zoning layer output by the nutrient monitoring module, and the lodging area layer output by the lodging detection module.

[0149] All received layers are spatially overlaid with the water and fertilizer management units in the standardized farmland spatial base map to determine the types of abnormal events contained in each water and fertilizer management unit and their spatial coverage.

[0150] For a water and fertilizer management unit that includes multiple types of abnormal events, the primary management problem of the unit is determined based on a preset agronomic treatment priority rule base. The priority rule base defines the sorting logic of high disease risk taking precedence over nutrient abnormality, nutrient abnormality taking precedence over irrigation abnormality, and irrigation abnormality taking precedence over weed risk.

[0151] For each water and fertilizer management unit, based on its primary management problem, the corresponding handling operation text is retrieved from the standardized operation instruction library. The handling operation text includes a problem location description and action instructions.

[0152] In this embodiment, the workstation opens a memory-based layer manager and loads the geographic information layer files output by the following external modules in sequence: the irrigation anomaly area layer output by the irrigation anomaly identification module, which contains leakage and blockage polygons; the seedling absence point layer and weed area layer output by the seedling anomaly identification module; the disease risk zoning layer output by the disease risk assessment module, which contains polygons of different risk levels; the nutrient anomaly zoning layer output by the nutrient monitoring module, which contains nutrient deficiency and excess polygons; and the lodging area layer output by the lodging detection module. At the same time, the workstation loads the standardized farmland spatial base map generated in step 5, which contains the boundary and unique identifier of each water and fertilizer management unit.

[0153] The workstation iterates through each water and fertilizer management unit in the standardized farmland spatial base map. For the current unit, the workstation creates an empty dictionary object to record various abnormal events occurring within that unit. Then, it sequentially performs spatial overlay analysis on the vector polygon of the current unit with each of the aforementioned abnormal layers. Specifically, for the missing seedling layer, the workstation calculates the number of missing seedlings falling within the boundary of the current unit. If the number is greater than 0, it records the missing seedling abnormality and the specific number of points in the dictionary. For the weed area layer, the workstation calculates the total area of ​​pixels covered by weeds within the current unit, divides it by the total area of ​​the unit, and obtains the weed coverage rate. If the coverage rate is greater than a preset threshold, such as 5%, it records the weed abnormality and the coverage rate percentage. For the disease risk zoning layer, the workstation identifies the overlapping part between the current unit and the high-risk disease area. If the overlapping area is greater than 0, it records the high disease risk and the overlapping area. For the nutrient anomaly zone layer, the workstation identifies the overlapping parts of the current cell with the nutrient deficiency area and the overlapping parts of the over-fertilization area, and records the nutrient deficiency anomaly and the nutrient over-fertilization anomaly, as well as their respective coverage areas; for the irrigation anomaly zone layer, the workstation identifies the overlapping parts of the current cell with the leakage area and the blockage area, and records the corresponding anomalies; for the lodging area layer, the workstation records the lodging anomaly and its area; for cells that do not overlap with any anomalies, the dictionary record is empty.

[0154] The workstation loads a pre-defined priority rule base, which is an ordered list. Items in the list are arranged from highest to lowest priority, specifically: lodging anomaly, high-risk disease, nutrient deficiency anomaly, irrigation leakage anomaly, irrigation blockage anomaly, nutrient excess anomaly, weed anomaly, and missing seedling anomaly. The workstation iterates through all anomaly types in the dictionary generated in step 2 for the current unit. For each anomaly type in the dictionary, the workstation finds its ranking number in the priority rule base; the smaller the number, the higher the priority. Then, it selects the anomaly type with the smallest ranking number, i.e., the highest priority anomaly type, as the unit's primary management issue. If the dictionary is empty, the primary management issue is marked as no anomaly.

[0155] The workstation maintains a standardized operation command library, which is a key-value mapping table. The key represents the exception type, and the value is a predefined Chinese text indicating the appropriate action. For each water and fertilizer management unit, the workstation searches the command library using the primary management problem type determined in step 3 as the key. For example, if the primary management problem is high-risk disease, the found text indicates a high-risk disease area, suggesting immediate application of a protective fungicide and focused inspection of the central diseased plants. If it's a nutrient deficiency anomaly, the text indicates a nutrient-deficient area, suggesting topdressing within three days, with the amount calculated based on the unit area. If it's an irrigation leak anomaly, the text indicates a suspected irrigation pipe leak, requesting maintenance personnel to bring ground-penetrating radar equipment to the marked point for verification. The workstation associates this text with the unit's unique identifier, forming a task entry.

[0156] Furthermore, the method also includes a visualization report generation step:

[0157] Create a blank geographic information system layout;

[0158] In the first area of ​​the layout, the standardized farmland spatial base map is loaded and displayed, and the primary management problem types of each water and fertilizer management unit are overlaid in a layered color scheme.

[0159] In the second area of ​​the layout, a list of agricultural tasks is generated in the form of a list. Each item in the list corresponds to a water and fertilizer management unit and includes the unit number, the type of primary management problem, and the corresponding handling operation text.

[0160] In the third area of ​​the layout, for the selected item in the agricultural task list, the original orthophoto local slice of the water and fertilizer management unit to which the unit belongs is dynamically loaded and displayed, and the specific spatial location of each abnormal event in the unit is marked with a highlighted boundary on the slice.

[0161] Output the agricultural monitoring report in a portable document format.

[0162] In this embodiment, the workstation calls the layout creation interface of the geographic information system component to generate a blank layout canvas with a standard A3 paper size. The layout canvas contains three predefined areas: the first area is located on the left side of the layout, occupying approximately 50% of the width; the second area is located on the upper right side of the layout, occupying approximately 30% of the width and 50% of the height; and the third area is located on the lower right side of the layout, occupying approximately 30% of the width and 50% of the height. The background of the layout is set to white, and the margin is set to 1 cm.

[0163] The workstation reads a standardized farmland spatial base map from memory and loads it as the base map layer into the first area of ​​the layout. Then, it iterates through all water and fertilizer management units. For each unit, based on its primary management problem type, it retrieves the corresponding fill color from a predefined color map table; for example, lodging anomalies are mapped to dark red, high disease risk to red, nutrient deficiency to orange, irrigation leakage to yellow, irrigation blockage to light yellow, nutrient excess to brown, weed anomalies to green, missing seedlings to light green, and no anomalies to gray. The workstation draws the vector polygon of each unit as its geometry using the corresponding fill color, while also drawing a black boundary line. Below the first area, a legend is added explaining the primary management problem type represented by each color. Additionally, a north arrow and scale bar are added to the upper right corner of the first area.

[0164] The workstation generates an agricultural task list based on the primary management issues of all water and fertilizer management units. This list is sorted according to a priority rule base, first listing all units with lodging anomalies, then listing high-risk diseases, nutrient deficiencies, etc. For each unit, the list includes the following information: unit number read from the standardized base map, primary management issue type, and treatment operation text obtained from step 4. The workstation formats this information into an HTML table with the unit number, issue type, and treatment suggestion as headers; this table is then rendered in the second area of ​​the layout. Simultaneously, a title "Agricultural Task List Sorted by Priority" is added above the first row of the table, and explanatory text is added below the table: "Please perform field operations in the order listed."

[0165] The workstation provides interactive functionality: when a user clicks on a row corresponding to a cell in the agricultural task list table in the second area of ​​the layout, the workstation captures the click event and obtains the cell number of the clicked row. Then, based on this cell number, the workstation locates the corresponding vector polygon from the standardized farmland spatial base map. Next, the workstation reads the coordinates of the smallest bounding rectangle of this polygon and crops the image sub-map within that rectangle from the orthophoto file generated in step 2. This sub-map is scaled to fit the size of the third area and rendered. Simultaneously, the workstation obtains the specific spatial locations of all abnormal events recorded in step 2 for this cell, such as the coordinates of missing seedlings, the boundary polygons of weed areas, and the boundary polygons of leak areas. These abnormal event graphics are overlaid on the image sub-map in a semi-transparent, highlighted form. For example, missing seedlings are marked with red dots, weed areas are covered with semi-transparent green, and leak areas are marked with blue outlines. At the bottom of the third area, a text label is added, displaying the cell number of the currently selected cell and a statistical summary of the types of abnormal events it contains.

[0166] The workstation provides a report export function. When the user clicks the "Export Report" button, the workstation rasterizes the current content of the first, second, and third layout areas into a single still image. Then, it uses a portable document format generation library to create a new document, inserting the aforementioned image as the first page. In subsequent pages of the document, the workstation appends thumbnails and metadata information for all original layers used in this analysis, including the orthophoto's capture time, spatial resolution, and drone model. Finally, the document is saved to the user-specified file path, named "Agricultural Monitoring Report_Current Date.pdf".

[0167] Furthermore, after generating the standardized farmland spatial base map in step S5, the process also includes an inspection path planning step:

[0168] Obtain all water and fertilizer management units identified as containing areas with irrigation anomalies, and extract the geographic centroid coordinates of each such unit;

[0169] Starting from the current location of the drone and taking the coordinates of the geographic centroid as the waypoints, a path search algorithm is used to calculate the shortest traversal path that passes through all waypoints, where the movement cost between two waypoints is determined by their Euclidean distance.

[0170] The shortest traversal path is converted into a sequence of waypoints that can be executed by the UAV, with each waypoint containing the target longitude, target latitude, and hovering observation commands.

[0171] In this embodiment, the workstation loads a generated irrigation anomaly area layer, which contains anomalous polygons marked as leaks or blockages. Simultaneously, a standardized farmland spatial base map is loaded. The workstation iterates through all irrigation anomaly polygons, and for each anomaly polygon, uses a spatial query function to find intersecting or contained water and fertilizer management units, recording their identifiers. To avoid duplication, the workstation creates a set, storing all found unit identifiers after deduplication. Then, for each water and fertilizer management unit in the set, the workstation reads the unit's vector polygon from the standardized farmland spatial base map and calculates its centroid coordinates. The centroid is calculated by averaging the longitude coordinates of all vertices of the polygon to obtain the centroid longitude, and averaging the latitude coordinates of all vertices to obtain the centroid latitude.

[0172] Step 2: Calculate the shortest traversal path that passes through all the waypoints.

[0173] The workstation obtains the latitude and longitude coordinates of the drone's current location as the starting point of the path. All centroid coordinates obtained in step 1 are used as a set of waypoints. The workstation employs a heuristic path search algorithm suitable for the Traveling Salesman Problem to plan the shortest traversal path. The specific implementation is as follows: First, create an unvisited waypoint list, initialized as a copy of all centroid coordinates. The starting point is set as the current point. Then, repeat the following steps until the unvisited list is empty: calculate the Euclidean distance from the current point to each point in the unvisited list, converting it to meters based on the latitude and longitude difference; select the closest point as the next target point; remove this point from the unvisited list; record the line segment between the current point and the next target point as part of the path; update the current point as the next target point. Finally, the algorithm terminates when all waypoints have been visited. The recorded sequence of line segments constitutes the shortest traversal path passing through all the centroids of the irrigation anomaly units. This is actually a greedy nearest neighbor heuristic solution, which does not guarantee global optimality but is computationally fast.

[0174] Step 3: Convert the shortest traversal path into a sequence of waypoints that can be executed by the UAV.

[0175] The workstation acquires the path segment sequence generated in step 2. For each segment in the path, its starting and ending points are converted into UAV waypoints. The definition of a waypoint includes the following information: target latitude, target longitude, and target altitude are set to fixed values, such as 50 meters above the ground, consistent with the observation altitude required for irrigation inspection, as well as the actions to be performed upon reaching the waypoint. For the first waypoint in the path, i.e., the first center of gravity after the UAV's starting point, the workstation adds the action of hovering for 10 seconds and taking multi-angle photos; for the last center of gravity in the path, the added action is to hover for 10 seconds and then automatically return to home; for the other center of gravity in between, the added action is to hover for 5 seconds and take orthogonal photos. In addition, the workstation adds a transition flight segment command between every two adjacent waypoints, which includes a flight speed of, for example, 8 meters per second and a straight flight mode. All waypoints are arranged in the order of visit, forming a complete waypoint sequence. The workstation converts the sequence into a standard UAV mission file format. This file uses a plain text structure, with each line representing a waypoint. Fields are separated by commas and are, in order, sequence number, latitude, longitude, altitude, and action command. Finally, the workstation uploads the mission file to the UAV's flight controller via a wireless data transmission link, allowing the UAV to automatically perform specialized inspection tasks in areas with irrigation anomalies.

[0176] According to a second embodiment of the present invention, the present invention claims protection for an integrated air-ground agricultural monitoring system based on intelligent field patrol, comprising:

[0177] One or more processors;

[0178] A memory that stores one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the aforementioned integrated air-ground agricultural monitoring method based on intelligent field patrol.

[0179] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0180] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0181] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A method for integrated air-ground agricultural monitoring based on intelligent field patrol, characterized in that, Includes the following steps: S1, acquire a sequence of UAV images of the target farmland, the sequence of images including visible light images and thermal infrared images; S2, Spatial registration and geometric correction are performed on the UAV image sequence to generate orthophotos and digital surface models covering the target farmland; S3, Based on the texture features of the orthophoto and the elevation change information of the digital surface model, extract the planting boundary of the farmland and identify the field ridges, roads and ditches inside the farmland; S4. Based on the planting boundary and the internal structure, and combined with the connected component analysis algorithm, the farmland area is divided into multiple water and fertilizer management units, each of which corresponds to an independent spatial vector polygon. S5. Based on the spatial vector polygon of the water and fertilizer management unit, a standardized farmland spatial base map is generated. The base map includes the boundary, area and spatial topology of each unit.

2. The method according to claim 1, characterized in that, S4 includes: Based on the pixel connectivity within the planting boundary, continuous crop planting areas are identified as initial units; For initial cells with an area lower than a preset threshold, an adjacent cell merging operation is performed, merging them into neighboring cells that share the longest common boundary; For initial units with an area greater than a preset threshold, they are divided based on vector lines of internal field ridges or ditches to generate multiple sub-units; Assign a unique identifier to each of the final generated water and fertilizer management units and record its geometric center coordinates and boundary length.

3. The method according to claim 1, characterized in that, After generating the orthophoto and digital surface model covering the target farmland in step S2, the process also includes an irrigation anomaly identification step: Obtain the canopy temperature distribution map corresponding to the thermal infrared image, and obtain the air temperature recorded by the synchronous meteorological station; Calculate the canopy temperature difference value for each pixel in the canopy temperature distribution map; Spatial continuity analysis was performed on the crown temperature difference values ​​to extract abnormal patches that form continuous closed regions; Based on the spatial morphological characteristics of the abnormal patches, abnormal patches that extend linearly along the irrigation pipeline are identified as suspected leakage areas, and abnormal patches that are isolated points or small clusters with a crown temperature difference higher than the surrounding environment are identified as suspected blockage areas. Output a risk distribution map that includes the geographical location and spatial extent of the suspected leak area and the suspected blockage area.

4. The method according to claim 1, characterized in that, After extracting the planting boundary of farmland based on the texture features of the orthophoto and the elevation change information of the digital surface model in step S3, the process also includes a seedling anomaly identification step. Acquire high-resolution visible light orthophotos of crop seedlings; An instance segmentation algorithm is performed on the high-resolution visible light orthophoto to identify individual crop pixel regions and background pixel regions in the image, wherein the background pixel regions include soil pixels and weed pixels. Based on the individual pixel regions of the crops, the geometric center point of each crop is extracted, and spatial clustering is performed on all center points to fit and generate the center line of the crop row. Along the center line of the crop row, the Euclidean distance between the center points of two adjacent crop individuals is calculated as the measured plant spacing, and the measured plant spacing is compared with the preset reference plant spacing. The positions where the measured plant spacing is greater than the upper limit of the reference plant spacing are marked as missing seedlings. Texture feature analysis is performed on the background pixel area. Areas with texture complexity lower than the preset weed threshold are identified as bare soil areas, and areas with texture complexity higher than or equal to the preset weed threshold are identified as weed areas. A weed risk distribution map is then generated.

5. The method according to claim 1, characterized in that, After generating the standardized farmland spatial base map in S5, the process also includes a growth period and disease risk assessment step: Acquire multi-temporal orthophotos of the target farmland at multiple consecutive time points, and calculate the wide dynamic range vegetation index map for each time point; Based on the wide dynamic vegetation index map, a time-series curve of vegetation index change over time is constructed within each water and fertilizer management unit. Extract the monotonically increasing intervals, peak points, and monotonically decreasing intervals from the time-series curve, and map the curve interval where the current time point is located to the seedling stage, jointing stage, heading stage, or maturity stage according to the preset phenological stage criteria. Acquire daily meteorological data of the target farmland, including daily maximum temperature, daily minimum temperature and relative humidity, and determine the host sensitivity coefficient based on the currently determined growth stage; Based on the daily meteorological data, the host sensitivity coefficient, and the preset disease occurrence conditions, the disease infection index of each water and fertilizer management unit is calculated, and the disease infection index is divided into low-risk, medium-risk, and high-risk levels.

6. The method according to claim 1, characterized in that, After generating the standardized farmland spatial base map in step S5, a nutrient monitoring step is also included: Acquire multi-band orthophotos of the target farmland, wherein the multi-bands include at least the near-infrared band and the red-edge band; Based on the reflectance values ​​of the near-infrared band and the red-edge band, the chlorophyll index value of each pixel is calculated to generate a nutrient status distribution map. Local spatial autocorrelation analysis was performed on the nutrient status distribution map to identify low-value clusters with nutrient index values ​​significantly lower than the neighborhood mean and high-value clusters with nutrient index values ​​significantly higher than the neighborhood mean. The low-value clustering areas are marked as potential nutrient deficiency areas, and the high-value clustering areas are marked as potential over-fertilization or fertilizer damage areas. The potential nutrient deficiency areas, potential over-fertilization or fertilizer damage areas are overlaid with the spatial vector polygons of the water and fertilizer management units to generate an independent list of nutrient anomalies for each unit.

7. The method according to claim 1, characterized in that, After generating the orthophoto and digital surface model covering the target farmland in step S2, the process also includes a lodging detection step: Acquire the first orthophoto and the first digital surface model in the pre-disaster phase, and the second orthophoto and the second digital surface model in the post-disaster phase; The first digital surface model and the second digital surface model are subjected to pixel-by-pixel difference calculation to generate a canopy height change map, wherein the area with a negative pixel value indicates a decrease in canopy height; Extract all consecutive pixel regions whose height decrease exceeds a preset threshold from the canopy height change map as candidate lodging regions; For each candidate collapsed area, calculate its first texture directionality measure on the first orthophoto and its second texture directionality measure on the second orthophoto. The region where the second texture directionality metric is significantly attenuated compared to the first texture directionality metric, and where the spatial overlap rate between the significantly attenuated region and the candidate collapsed region exceeds a preset overlap threshold, is ultimately determined to be a confirmed collapsed region. The total number of pixels in the confirmed lodging area is counted, and the lodging area is calculated by combining the spatial resolution of the second orthophoto. Based on the average height reduction value of the canopy height change map in the confirmed lodging area, the lodging is classified into mild, moderate, and severe levels.

8. The method according to claim 1, characterized in that, Following S5, the process also includes a step for generating agricultural information recommendations: Receives the irrigation anomaly area layer output by the irrigation anomaly identification module, the seedling missing point and weed area layer output by the seedling anomaly identification module, the disease risk zoning layer output by the disease risk assessment module, the nutrient anomaly zoning layer output by the nutrient monitoring module, and the lodging area layer output by the lodging detection module. All received layers are spatially overlaid with the water and fertilizer management units in the standardized farmland spatial base map to determine the types of abnormal events contained in each water and fertilizer management unit and their spatial coverage. For a water and fertilizer management unit that includes multiple types of abnormal events, the primary management problem of the unit is determined based on a preset agronomic treatment priority rule base. The priority rule base defines the sorting logic of high disease risk taking precedence over nutrient abnormality, nutrient abnormality taking precedence over irrigation abnormality, and irrigation abnormality taking precedence over weed risk. For each water and fertilizer management unit, based on its primary management problem, the corresponding handling operation text is retrieved from the standardized operation instruction library. The handling operation text includes a problem location description and action instructions.

9. The method according to claim 8, characterized in that, It also includes the step of generating a visual report: Create a blank geographic information system layout; In the first area of ​​the layout, the standardized farmland spatial base map is loaded and displayed, and the primary management problem types of each water and fertilizer management unit are overlaid in a layered color scheme. In the second area of ​​the layout, a list of agricultural tasks is generated in the form of a list. Each item in the list corresponds to a water and fertilizer management unit and includes the unit number, the type of primary management problem, and the corresponding handling operation text. In the third area of ​​the layout, for the selected item in the agricultural task list, the original orthophoto local slice of the water and fertilizer management unit to which the unit belongs is dynamically loaded and displayed, and the specific spatial location of each abnormal event in the unit is marked with a highlighted boundary on the slice. Output the agricultural monitoring report in a portable document format.

10. A sky-ground integrated agricultural monitoring system based on intelligent field patrol, characterized in that: include: One or more processors; A memory having stored one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the integrated air-ground agricultural monitoring method based on intelligent field patrol as described in any one of claims 1 to 9.

Images