Crop growth monitoring method based on unmanned aerial vehicle routing inspection route planning aerial photography data

Abstract

The invention discloses a crop growth monitoring method based on unmanned aerial vehicle routing inspection route planning aerial data. The method comprises the following steps: segmenting and screening out a polygonal farmland area in a satellite map; sampling parameters are set in the unmanned aerial vehicle inspection planning model, the flight height and the route spacing of the unmanned aerial vehicle are configured based on the sampling parameters, and the unmanned aerial vehicle inspection planning model plans and outputs a one-way flight inspection route corresponding to the polygonal farmland area; the unmanned aerial vehicle performs inspection flight in the polygonal farmland area to shoot farmland panoramic image data of the polygonal farmland area; and constructing a crop growth monitoring analysis model, and inputting the farmland panoramic image data into the crop growth monitoring analysis model for crop type identification and segmentation, growth stage affiliation identification marking and abnormity identification marking processing to obtain a crop growth monitoring result. According to the invention, crop types, growth stages and abnormal information can be displayed at corresponding positions of the farmland panoramic image, and an intelligent agricultural monitoring report is formed.

Description

Technical Field

[0001] This invention relates to the field of agricultural crop monitoring, and in particular to a method for monitoring crop growth based on aerial photography data from drone inspection route planning. Background Technology

[0002] Automated identification and inspection of large areas of farmland are the cornerstone of modern smart farmland. Currently, large-scale agricultural crop monitoring primarily relies on satellite remote sensing technology. However, satellite remote sensing is affected by factors such as sensor physical characteristics, electromagnetic wave propagation laws, and environmental interference. For example, clouds, fog, and aerosols (such as haze and dust) absorb / scatter solar radiation, causing optical satellite images to fail to capture crop spectral signals or suffer from radiation distortion (spectral signal distortion). Consequently, the acquired satellite spectral images of agricultural farmland areas suffer from low quality and a low proportion of effective data. Currently, satellite spectral imagery is mainly used for macroscopic identification and boundary delineation of large-scale farmland areas, failing to achieve refined monitoring of farmland regions. It cannot effectively and precisely monitor crop growth conditions such as crop type, growth stage, and abnormal growth (e.g., pests and diseases), thus failing to achieve the goal of large-scale crop growth monitoring, especially in complex farmland areas where growth changes are difficult to accurately capture. Currently, crop growth monitoring mainly relies on on-site inspections and recordings by professionals. However, manual inspections are not only time-consuming and labor-intensive but also have low farmland coverage, low monitoring efficiency, and are susceptible to human error due to a lack of standardized procedures. Summary of the Invention

[0003] The purpose of this invention is to provide a method for monitoring crop growth based on aerial photography data from drone inspection route planning. This method combines manual inspection with a ground feature classification model to segment polygonal farmland areas. A drone inspection planning model is used to plan and design unidirectional flight routes corresponding to these complex polygonal farmland areas. The drone then conducts inspections and captures images along these unidirectional routes at the specified altitudes, creating panoramic images of the farmland. These images are then processed using a trained crop growth monitoring and analysis model to identify and segment crop types, assign growth stages, and identify anomalies. Finally, the method outputs the crop growth monitoring results. The method can display crop types, growth stages, and anomaly information at corresponding locations in the panoramic farmland images and generate a smart agriculture monitoring report.

[0004] The objective of this invention is achieved through the following technical solution: A method for monitoring crop growth based on aerial photography data from drone inspection route planning, the method comprising: S1. Segment and filter out polygonal farmland areas in the satellite map; S2. Construct a UAV inspection planning model. In the UAV inspection planning model, set the sampling parameters of the polygonal farmland area and configure the flight altitude and flight path spacing of the UAV based on the sampling parameters. The UAV inspection planning model plans and outputs the one-way flight inspection route corresponding to the polygonal farmland area. S3. The drone takes pictures of the polygonal farmland area by flying at the flight altitude and one-way flight inspection route. The pictures are then stitched together to obtain panoramic image data of the polygonal farmland area. S4. Construct a crop growth monitoring and analysis model and a sample dataset. The sample dataset is constructed according to the structural hierarchy of crop type, growth cycle, and sample image. The growth cycle is divided into several growth stages according to crop type. Corresponding abnormal label information is marked on the sample images. The abnormal label information includes the abnormality type and the severity of the abnormality. The crop growth monitoring and analysis model is trained using the sample dataset. The panoramic image data of farmland is input into the crop growth monitoring and analysis model for crop type identification and segmentation, growth stage identification and labeling, and abnormal identification and labeling processing. Finally, the crop growth monitoring results are output.

[0005] To better implement this invention, the crop growth monitoring results include a panoramic image feature map of farmland and statistical analysis data. The panoramic image feature map of farmland displays the crop type, growth stage, and abnormal information at the corresponding location in the panoramic image of farmland. The abnormal information includes the abnormal location, abnormal type, and abnormal severity in the panoramic image feature map of farmland. The statistical analysis data includes statistical data on crop types, statistical data on the classification of crop types corresponding to growth stages, and statistical data on abnormal areas.

[0006] Preferably, in method S1, satellite image data is acquired and input into a satellite map containing latitude and longitude information; a land cover classification model for the segmentation of farmland target areas is constructed; the land cover classification model performs farmland target area identification and segmentation processing on the satellite image data, and obtains several polygonal farmland areas in the satellite image; and the polygonal farmland areas to be studied are selected.

[0007] Preferably, the method for obtaining a one-way flight inspection route using the UAV inspection planning model includes: S21. Obtain the polygon vertex set of the polygon farmland region boundary and perform centering and covariance matrix calculation, then obtain the eigenvalues ​​of the covariance matrix. With the corresponding first feature vector Second eigenvector ; with the first eigenvector The main direction, with the second feature vector A secondary direction perpendicular to the primary direction; S22. Select the vertex with the longest vertical distance in the main direction as the starting point, and generate several parallel lines parallel to the main direction along the secondary direction according to the flight path spacing; obtain the intersection points of all parallel lines with the boundary of the polygonal farmland area. S23. Using parallel lines as the route segments, starting from the starting point, the parallel lines are sequentially numbered according to their secondary directions; starting from the starting point, each intersection point is sequentially connected by a one-way flight route according to the parallel line number to obtain a set of one-way flight route paths. All route path sets constitute a one-way flight inspection route.

[0008] Preferably, in method S2, the drone is equipped with a camera, and the sampling parameters include the ground sampling distance of the images captured by the camera, the image overlap rate, and the shooting tilt angle; the drone's flight altitude. Calculated using the following formula: ,in To convert the tilt angle of the shot into a vertical downward shooting distance, For pixel width, For the width of the camera sensor, The focal length of the camera; Flight path spacing of drones Calculated using the following formula: , This represents the image overlap rate.

[0009] Preferably, in method S21, several sampling points are uniformly interpolated on each edge of the polygon vertex set to form an interpolation point set containing vertices and sampling points. The mean of the interpolation point set is then centered as the centroid. Then, a description is constructed. , The eigenvalues ​​are obtained by eigenvalue decomposition of the 2×2 covariance matrix of directional correlation. With the corresponding first feature vector Second eigenvector .

[0010] Preferably, in method S22, the neighboring edge vectors associated with the starting point of the selected polygonal farmland region are extracted. If the neighboring edge is related to the first feature vector... If the included angle between them is greater than or equal to Q1°, then the principal direction is changed to the adjacent edge vector, and the secondary direction is changed to the direction perpendicular to the principal direction; if the adjacent edge vector and the first eigenvector are... If the included angle is less than Q1°, no changes or adjustments will be made.

[0011] Preferably, in method S22, a latitude and longitude coordinate system is constructed in the satellite map, and the polygonal farmland area, starting point, parallel lines, and intersection points are expressed in the latitude and longitude coordinate system; in method S23, all path segments are collected in the latitude and longitude coordinate system, and the path segments are sequentially encoded according to the secondary direction starting from the starting point. The path segment with the first ranking is selected and the intersection point closest to the starting point is extracted as the second point of the flight path. The starting point is connected to the second point, and another intersection point of the first ranking path segment is extracted as the third point of the flight path. The intersection point closest to the third point of the second ranking path segment is selected as the fourth point of the flight path. The third point and the fourth point of the flight path are connected, and so on, sequentially connecting each intersection point according to the path segment number.

[0012] Preferably, in method S22, the parallel lines are all parallel lines that intersect the boundary of the polygonal farmland area, and parallel lines that exceed the boundary of the polygonal farmland area are deleted.

[0013] Preferably, in method S23, the connecting paths and all vertices between adjacent path segments are filtered. If there is a vertex whose distance to the connecting path is greater than the route spacing, the connecting path is split into two segmented paths with that vertex as the path point. If the distance between the vertex and the connecting path is less than or equal to the route spacing, no processing is performed.

[0014] Compared with the prior art, the present invention has the following advantages and beneficial effects: (1) This invention combines manual inspection and land feature classification model to identify and segment polygonal farmland areas. It plans and designs one-way flight inspection routes corresponding to complex polygonal farmland areas through UAV inspection planning model. The UAV is allowed to inspect and photograph the polygonal farmland areas according to the flight altitude and one-way flight inspection route to obtain the panoramic image data of the farmland corresponding to the polygonal farmland areas. Then, the crop growth monitoring and analysis model is trained to identify and segment crop types, identify and mark growth stages, and mark anomalies, and finally output the crop growth monitoring results. It can display the crop type, growth stage and anomaly information at the corresponding position of the panoramic image of the farmland and form a smart agriculture monitoring report.

[0015] (2) This invention takes "regional definition, intelligent identification, automatic inspection, and analysis output" as its core, and can regularly perform drone automated inspection and aerial photography tasks on complex polygonal farmland areas, realize intelligent monitoring of farmland throughout the entire cycle and regularly generate smart agriculture monitoring reports; through full-coverage drone automated inspection without blind spots, it makes up for the technical shortcomings of the extensive use of satellite remote sensing data and the time-consuming and labor-intensive nature of manual inspection, and improves the automation level of farmland identification and crop anomaly detection under complex terrain; at the same time, it supports the full-process automated execution of custom cycles and comparison with historical data trends, which reduces the time of the entire process from data collection to analysis, not only reducing the labor cost of maintenance, but also increasing the frequency of drone inspections, promoting the transformation of farmland monitoring from passive early warning to active prevention and control, and improving the practicality and flexibility of farmland management.

[0016] (3) The UAV inspection planning model of the present invention first receives the user input of the overlap rate, camera parameters and other configurations. After parameter verification and completion, it loads or generates a polygon point set of the target area. Then, it determines the optimal flight direction through PCA analysis and adjacent side angle judgment, generates parallel routes and calculates the intersection of the routes and polygons. Then, starting from the initial point, it generates a complete path according to the rule of finding the shortest path along the boundary between groups and connecting matching points along parallel lines within groups. At the same time, it deduplicates and optimizes. Finally, it saves the path in a JSON file, visualizes the planning results, and outputs statistical information such as flight altitude and number of routes. It can generate efficient and precise cruise routes covering polygon farmland areas for UAVs and provide accurate path guidance for UAV inspection autonomous operations. Attached Figure Description

[0017] Figure 1 This is a flowchart of the crop growth monitoring method of the present invention; Figure 2 This is a flowchart illustrating the method for planning a one-way flight inspection route using a drone inspection planning model in this embodiment. Figure 3 This is a schematic diagram illustrating the generation of parallel lines along a secondary direction in a polygonal farmland area according to the flight path spacing, as exemplified in the embodiment. Figure 4 This is a schematic diagram illustrating the principle of obtaining a set of flight path combinations from another polygonal farmland area, as exemplified in the embodiment. Detailed Implementation

[0018] The present invention will be further described in detail below with reference to embodiments: Example like Figure 1 As shown, a method for monitoring crop growth based on aerial photography data from drone inspection route planning includes the following steps: S1. Polygonal farmland areas are segmented and selected from satellite maps. In practice, several farmland areas are segmented from the satellite map using information from manual field inspections. Then, the selected polygonal farmland areas are used to monitor crop growth. In some embodiments, this invention can also acquire satellite imagery and input it into a land cover classification model for large-area farmland area identification and segmentation. The method is as follows: Satellite remote sensing image data is acquired and input into a satellite map containing latitude and longitude information; a land cover classification model focusing on the segmentation of farmland target areas is constructed. The land cover classification model mainly identifies farmland areas from large areas. For example, this invention is applied to a suburban area of ​​a city. Satellite remote sensing image data of a large geographical area of ​​the suburban area is acquired. The land cover classification model identifies and segments farmland areas (including irregular polygonal boundaries of farmland areas) based on the satellite remote sensing image data. The land cover classification model can be built using U-Net, DeepLab, Transformer, SAM, etc. The main purpose of the land cover classification model in this embodiment is to identify and segment farmland areas, focusing on the segmentation of farmland target areas. The land cover classification model identifies and segments farmland target areas from satellite imagery data. The model's sample dataset includes positive and negative farmland samples. Positive samples include those with various single crops and those with combinations of multiple crops. Negative samples include samples of buildings, woodlands, water bodies, and roads. To improve the accuracy of farmland segmentation, the negative samples include many similar samples such as woodlands, grasslands, and gardens. The model segments the satellite imagery into several polygonal farmland areas, and then selects the polygonal farmland areas for study.

[0019] In some embodiments, during the peak growing season for crops in farmland areas, a rough target farmland area is selected from satellite maps based on information from manual field inspections. Then, the latest satellite imagery data of the target farmland area is acquired (based on regional coordinates and preset rules such as resolution, cloud cover, and time, the optimal image is selected, georegistration and data import are completed, and satellite remote sensing imagery data of the target farmland area is generated). Image preprocessing is performed on the satellite remote sensing imagery data, using it to calculate the Normalized Difference Vegetation Index (NDVI), Spectral Index (EVI), and SAVI for each location point (pixel or cell). A land cover classification model is used to perform more detailed identification and segmentation of the target farmland area, resulting in farmland extraction results with coordinate information. An appropriate number of boundary vertices are extracted from the farmland extraction results to obtain several polygonal farmland areas. All polygonal farmland areas can be converted into vector data in geojson format (containing the boundary vertices coordinate information of farmland plots; each farmland plot is represented by a polygon composed of several vertices). These polygonal farmland areas are selected as the study area and serve as the core area for subsequent UAV inspections and crop analysis.

[0020] S2. Construct a UAV inspection planning model. In this model, set sampling parameters for the polygonal farmland area and configure the UAV's flight altitude and flight path spacing based on these parameters. The UAV is equipped with a camera to inspect and photograph the farmland area. Sampling parameters include the ground sampling distance (GSD) of the images captured by the camera (GSD refers to the actual distance of a pixel on the ground, i.e., the distance between the centers of two adjacent pixels), image overlap rate, and shooting tilt angle. The UAV's flight altitude... Calculated using the following formula: ,in To convert the tilt angle of the shot into a vertical downward shooting distance, It refers to the pixel width (i.e., the pixel width of the camera sensor, the number of pixels corresponding to the sensor width). This refers to the width of the camera sensor (i.e., the physical width of the camera sensor). This is the camera focal length. If the shooting tilt angle is... Then the ground sampling distance at the tilt angle of the image. Converted to ground sampling distance for vertical downward shooting The conversion expression is as follows: .

[0021] Flight path spacing of drones Calculated using the following formula: , This represents the image overlap rate. The UAV inspection planning model outputs a one-way flight inspection route corresponding to a polygonal farmland area.

[0022] In some embodiments, such as Figure 2 As shown, the method for obtaining a one-way flight inspection route using the UAV inspection planning model includes: S21. Obtain the polygon vertex set of the polygon farmland region boundary and perform centering and covariance matrix calculation. In some embodiments, before centering, the present invention uniformly interpolates several sampling points on each edge of the polygon vertex set to form an interpolation point set containing vertices and sampling points (preferably, the interpolation point set is used instead of the polygon vertex set for centering and covariance matrix calculation). The mean of the interpolation point set is obtained as the centroid by centering the interpolation point set. Eigenvalues ​​are obtained from the covariance matrix. With the corresponding first feature vector Second eigenvector Preferably, the present invention is constructed as described. , The eigenvalues ​​are obtained by eigenvalue decomposition of the 2×2 covariance matrix of directional correlation. With the corresponding first feature vector Second eigenvector Using the first eigenvector The main direction, with the second feature vector This refers to secondary directions perpendicular to the primary direction. Key features of the polygon (primary direction, initial point, parallel lines, intersections, etc.) are extracted using linear algebra, Euclidean geometry, and computational geometry methods to provide a precise geometric foundation for subsequent path generation. The core objective is to achieve "efficient coverage of the polygonal region." In some embodiments, to improve the accuracy of subsequent primary direction PCA extraction, sampling points need to be uniformly inserted on each edge of the polygon; let the polygon's... The starting point of the strip is The destination is (in a closed polygon) There are a total of n edges, and the interpolation formula is:

[0023] in This represents the total number of sampling points for each edge (default 20). For the first The first edge Several interpolation points are used to transform the discrete polygon boundary into a continuous point set through dense sampling, providing sufficient data support for subsequent main direction analysis. For the polygon vertex set, several sampling points are uniformly interpolated on each edge of the polygon to form an interpolation point set containing vertices and sampling points. The mean of the interpolation point set is then centered as its centroid. If the interpolation point set... ( (Total number of sampling points) The interpolation point set is centered, as shown in the following expression: in It is the mean (also known as the centroid) of the interpolation point set. The points after centering point k (each point minus the centroid, so that the data is distributed around the centroid).

[0024] Then, the covariance matrix is ​​calculated, and a 2×2 covariance matrix is ​​constructed to describe... , Directional relevance: ;in , Let be the variance in the x-direction; , Let be the covariance in the x and y directions. For the covariance matrix... Perform eigenvalue decomposition to obtain eigenvalues. With the corresponding feature vector First eigenvector : Principal direction (the direction in which the polygon extends the longest; the principal direction is defined as the "longest extension direction of the polygon," which determines the overall direction of the flight path); Second eigenvector Secondary direction (perpendicular to the primary direction, used to generate parallel lines).

[0025] In some embodiments, the vertex with the farthest vertical distance in the main direction is selected as the starting point (ensuring the takeoff point is located at the edge of the region to reduce invalid flights). The neighboring edge vectors associated with the starting point are extracted from the selected starting point in the polygonal farmland region. To improve the overall consistency of the effect, this invention corrects the main direction by judging the included angle of the neighboring edges. The specific method is as follows: If the adjacent edge vector and the first feature vector If the included angle between the two directions is greater than or equal to Q1° (10° in this example), the primary direction is changed to the adjacent vector (effectively avoiding excessive tilting of the flight path and boundary, improving flight safety), and the secondary direction is changed to a direction perpendicular to the primary direction; the adjacent vector and the first characteristic vector The angle between them is calculated using the dot product of vectors, and is expressed as follows: , The neighboring edge vector of the starting point, The neighboring edge vector and the first eigenvector The angle between the adjacent edge vector and the first eigenvector. If the included angle between them is less than Q1° (10° in this example), no changes or adjustments will be made.

[0026] S22. Select the vertex with the longest vertical distance in the main direction as the starting point, and generate several parallel lines parallel to the main direction along the secondary directions according to the flight path spacing (preferably, the parallel lines are all parallel lines that intersect the boundary of the polygonal farmland area; parallel lines that exceed the boundary of the polygonal farmland area are deleted). Figure 3 As shown, with Figure 3 Taking a polygonal farmland area as an example, the parallel lines drawn within the polygonal farmland area (see example) Figure 3 (blue line) Figure 3 As shown. In some embodiments, a latitude and longitude coordinate system is constructed to represent polygonal farmland areas, starting points, parallel lines, and intersection points.

[0027] In some embodiments, several equally spaced parallel lines (with a spacing equal to the flight path spacing) are generated along the secondary direction to cover the entire polygonal region. The projection range is then calculated, and all vertices of the polygon are projected onto the secondary direction (i.e., the second feature direction). On the projection range, the projection value range is obtained. The projection formula is: in It is a unit vector. Determine the position of the first line. Projection value from the starting point Offset the polygon inwards by half a step, ensuring coverage of the region at the initial point (i.e., the starting point); then iteratively generate parallel lines: along the secondary direction by step... (i.e., the spacing between flight paths) Iteratively generate parallel lines until they exceed the projection range. . No. The projection value of the line is: ;in The direction is determined by the position of the initial point (i.e., the starting point). For easier representation, two points are used. , , The route consists of segments express.

[0028] Obtain all intersections of parallel lines with the boundary of the polygonal farmland region. In some embodiments, the polygonal boundary consists of line segments connecting consecutive vertices, the first... A boundary line segment is defined as (in (ensure the polygon is closed).

[0029] The parametric equation of the line segment is: , ;in: , The two endpoints of the line segment are t and t is the parameter, representing the vector difference. - =(Δx,Δy), where Δx= - Δy= - Let the equation of the m-th route (parallel line) be... : , and boundary line segment The steps for calculating the intersection point are as follows: (1) Substitute into the parametric equation: Substitute into the flight path equation .

[0030] (2) Solving for parameters When the denominator is 0, the line segment is parallel to the flight path and there is no intersection.

[0031] (3) Intersection validity judgment: If t∈[0,1], then the coordinates of the intersection point are: Otherwise, the line segment and the flight path will not intersect.

[0032] Then, intersection point deduplication and sorting are performed, using the following methods: (1) Deduplication: Distance less than If they are considered to be the same point, then one of them will be retained.

[0033] (2) Sorting rule: along the main direction vector The projection values ​​are sorted from smallest to largest to ensure the drone flies unidirectionally along the flight path. The projection value of the intersection point p=(x,y) is: ,according to Sort in ascending order to obtain an ordered list of intersections within the flight path.

[0034] Sequence creation methods include the following: 1. Complete Sequence: Contains all vertices and intersections, arranged in clockwise boundary traversal order. Used for inter-group turning path calculation. That is, starting from the initial point, traversing the polygon clockwise, storing all encountered vertices and intersections in the order they are encountered. Starting from the initial point, traversing each edge of the polygon clockwise, the edge order is as follows... ,in As vertices arrive The edge, For vertices arrive Each edge : (1) Add the starting point (vertex) of the edge: add the vertex The mark With coordinates join sequence; (2) Add intersections on edges: Calculate all intersections on each edge to the starting point. Distance: Arrange the intersection points in ascending order of distance (from the starting point to the ending point), and add the sorted intersection points to the sequence in turn. (3) Closed sequence: After traversing all edges, add an initial point (vertex). This allows the sequence to form a closed loop (ensuring the drone can return to its starting point).

[0035] 2. Traversal Sequence: A simplified version of the complete sequence, containing only the initial point and intersections. It is used to quickly generate the core path of the route. That is, starting from the initial point, it traverses the polygon clockwise and stores all the intersections encountered in the order they are encountered. It retains the points whose first element in the marker is 0 (initial point, marker (0,0)) or m≥1 (intersection). The order of retention is consistent with the complete sequence to ensure the continuity of the route direction.

[0036] S23. Using parallel lines as flight paths, the parallel lines are sequentially numbered according to their secondary directions, starting from the origin. Starting from the origin, a unidirectional flight path is used to sequentially connect the intersection points according to the parallel line numbers, resulting in a set of unidirectional flight path lines. Preferably, continuous flight paths and corresponding segments within the continuous flight path set are obtained (the latitude and longitude data of the continuous flight paths are obtained for use in UAV-planned flight paths and UAV inspection navigation). Each segment includes path points, and path points include intersections and vertices. Both continuous flight paths and segments include latitude and longitude data. Preferably, all path segments are collected in a latitude and longitude coordinate system. Starting from the origin, each path segment is sequentially numbered according to its secondary direction. The first-ranked path segment is selected, and its closest intersection with the origin is extracted as the second point of the flight path. The origin is then connected to the second point, and another intersection of the first-ranked path segment is extracted as the third point of the flight path. The second-ranked path segment is selected, and its closest intersection with the third point is selected as the fourth point of the flight path. The third and fourth points are then connected. The third-ranked path segment is selected, and its closest intersection with the fourth point is selected as the fifth point of the flight path. The fifth and sixth points are then connected, and so on, sequentially connecting each intersection according to the path segment's number. For example... Figure 4 As shown, Figure 4The example shows a polygonal farmland area. The method of this invention yields a set of flight paths as shown in the blue trajectory. When m=0, the mark (m,n) indicates that the point is the nth inflection point encountered when walking counterclockwise along the boundary of the polygon from the starting point. The initial point is marked as (0,0). When m is greater than or equal to 1, it indicates that the nth intersection point is the nth parallel line sorted along the main direction starting from the mth parallel line closest to the initial point. The sequence of UAV waypoints (UAV waypoints are path points, including intersections and / or vertices) in the flight path set is as follows: Point 0: Marked (0, 0); Point 1: Marked (1, 1); Point 2: Marked (1, 2); Point 3: Marked (2, 2); Point 4: Marked (2, 1); Point 5: Marked (3, 1); Point 6: Marked (3, 2); Point 7: Marked (4, 2); Point 8: Marked (4, 1); Point 9: Marked (5, 1); Point 10: Marked (5, 2); Point 11: Marked (6, 2); Point 12: Marked (6, 1); Point 13: Marked (0, 2); Point 14: Marked (7, 1); Point 15: Marked (7, 2); Point 16: Marked (8, 2); Point 17: Point 18: Marked (9, 1); Point 19: Marked (9, 2); Point 20: Marked (10, 2); Point 21: Marked (10, 1). The optimized path data is converted into a command format recognizable by the UAV, such as a sequence of waypoint coordinates. Simultaneously, path visualization charts are generated, including the distribution of flight routes over the covered area and markers of key turning points, allowing users to intuitively view the path planning effect. The results are saved as JSON, supporting subsequent import into the UAV control system for flight mission execution.

[0037] In complex polygonal farmland applications, several parallel lines parallel to the main direction are generated along the secondary direction at flight path intervals. Some of these parallel lines intersect the boundary of the polygonal farmland area more than two times (i.e., there is a gap in the middle of the polygonal farmland area boundary that is recessed towards the center, and the UAV flight path cannot cross this gap). To address this, this invention determines the number of path segments, intersections, or areas on either side of the recessed gap. The side with a relatively larger number of path segments, intersections, or area (relative to the side of the recessed gap) is designated as the first-priority flight area, and the other side as the second-priority flight area. The unidirectional flight path sequentially connects the intersections of the first-priority flight areas according to their order of priority, and then connects the intersections of the second-priority flight areas. If there are multiple gaps in the polygonal farmland area that are recessed towards the center, the side with a relatively larger number of path segments, intersections, or area is selected as the first-priority flight area according to the adjacent order of the gaps. The order of the flight areas is then ordered according to the adjacent order of the gaps, and the unidirectional flight path sequentially connects the intersections of the flight areas according to their order of priority. The method of this invention yields a set of flight paths, as shown in the blue trajectory. The marker (m,n) indicates that when m=0, the point is the nth inflection point encountered while traversing counter-clockwise along the polygon boundary from the starting point; the initial point is marked as (0,0). When m is greater than or equal to 1, it indicates the nth intersection point along the main direction on the mth parallel line starting from the line closest to the initial point. In some embodiments, connecting paths and all vertices between adjacent path segments are filtered. If the distance between a vertex and the connecting path is greater than the flight path spacing, the connecting path is split into two segmented paths using that vertex as the path point. Vertexes whose distance from the connecting path is less than or equal to the flight path spacing are not processed.

[0038] S3. The UAV takes pictures of the polygonal farmland area by flying at the specified altitude and along a one-way flight path. The pictures are then stitched together to obtain the panoramic image data of the polygonal farmland area (i.e., the panoramic orthophoto of the farmland).

[0039] S4. Construct a crop growth monitoring and analysis model and a sample dataset. The sample dataset is constructed in sequence according to the structure hierarchy of crop type, growth cycle, and sample image. The crop type is divided according to the crop name, such as wheat, corn, rice, soybean, etc. The growth cycle is divided into several stages according to crop type. The commonly divided growth stages are: emergence (until seedling emerges), vegetative growth (from seedling emergence to rapid stem and leaf growth), reproductive growth (from flowering to grain and fruit formation), and maturity / harvest (crop full maturity). However, the growth cycle can also be divided into stages specific to certain crop types. For example, for annual summer corn, the growth stages are: emergence (from seed germination to seedling emergence, approximately mid-to-late June), jointing (when the number of leaves increases to a certain extent and the stem elongates, approximately early to late July), trumpet stage (when the central leaves unfold into a trumpet shape, approximately early August), flowering (when the tassel emerges, approximately mid-to-late August), grain-filling stage (when grains fill and the canopy begins to age, approximately early to late September), and maturity stage (when grains mature and the stem turns yellow, approximately October). (Early May); For example, for fruits and vegetables, the growth cycle is divided into the seedling stage, seedling establishment stage, flowering and fruit setting stage, fruit enlargement stage, and maturity stage. Anomaly labels are marked on the sample images, including the anomaly type and severity. Anomaly types include pest and disease types, nutrient deficiency types, and weed growth types. The coordinates of the anomaly area, severity (divided into low, medium, and high levels), and affected area are also marked. The crop growth monitoring and analysis model is trained using the sample dataset. Panoramic farmland image data is input into the crop growth monitoring and analysis model for crop type identification and segmentation, growth stage attribution identification and labeling, and anomaly identification and labeling processing, ultimately outputting crop growth monitoring results. Regular monitoring of crop growth in polygonal farmland areas is performed, yielding monitoring results and structured monitoring reports. The structured monitoring reports include basic regional information (location, farmland area, crop type), growth analysis (overall rating, plot comparison), anomaly warnings (textual and image annotations of anomaly areas), and management recommendations.

[0040] The crop growth monitoring results in this embodiment include a panoramic image feature map of the farmland and statistical analysis data. The panoramic image feature map displays the crop type, growth stage, and abnormal information at the corresponding location in the panoramic image. The abnormal information includes the abnormal location, abnormal type, and abnormal severity in the panoramic image feature map. The statistical analysis data includes statistical data on crop types, statistical data on the classification of crop types corresponding to their growth stages, and statistical data on abnormal areas. The statistical data on crop types includes the crop type (i.e., the crop name) and its area percentage. The statistical data on the classification of crop types corresponding to their growth stages is the percentage of crops classified according to their growth stages (i.e., the percentage of crops classified by growth stage). The statistical data on abnormal areas includes the abnormal type, abnormal severity, and abnormal area percentage marked at the corresponding location in the panoramic image. In some embodiments, a crop rating standard system is constructed. The statistical analysis data is used to evaluate the crop type according to the crop rating standard system (the crop rating standard system is a crop growth rating constructed according to crop type and growth stage hierarchy, with ratings divided into three levels: excellent, medium, and poor) to obtain the rating data corresponding to the crop type.

[0041] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring crop growth based on aerial photography data from drone inspection route planning, characterized in that: The methods include: S1. Segment and filter out polygonal farmland areas in the satellite map; S2. Construct a UAV inspection planning model. In the UAV inspection planning model, set the sampling parameters of the polygonal farmland area and configure the flight altitude and flight path spacing of the UAV based on the sampling parameters. The UAV inspection planning model plans and outputs the one-way flight inspection route corresponding to the polygonal farmland area. S3. The drone takes pictures of the polygonal farmland area by flying at the flight altitude and one-way flight inspection route. The pictures are then stitched together to obtain panoramic image data of the polygonal farmland area. S4. Construct a crop growth monitoring and analysis model and a sample dataset. The sample dataset is constructed according to the structural hierarchy of crop type, growth cycle, and sample image. The growth cycle is divided into several growth stages according to crop type. Corresponding abnormal label information is marked on the sample images. The abnormal label information includes the abnormality type and the severity of the abnormality. The crop growth monitoring and analysis model is trained using the sample dataset. The panoramic image data of farmland is input into the crop growth monitoring and analysis model for crop type identification and segmentation, growth stage identification and labeling, and abnormal identification and labeling processing. Finally, the crop growth monitoring results are output.

2. The method for monitoring crop growth based on aerial photography data from UAV inspection route planning according to claim 1, characterized in that: The crop growth monitoring results include a panoramic image feature map of the farmland and statistical analysis data. The panoramic image feature map displays the crop type, growth stage, and abnormal information at the corresponding location in the panoramic image of the farmland. The abnormal information includes the abnormal location, abnormal type, and abnormal severity in the panoramic image feature map of the farmland. The statistical analysis data includes statistical data on crop types, statistical data on the classification of crop types and corresponding growth stages, and statistical data on abnormal areas.

3. The method for monitoring crop growth based on aerial photography data from UAV inspection route planning according to claim 1, characterized in that: In method S1, satellite image data is acquired and input into a satellite map containing latitude and longitude information; a land cover classification model focusing on the segmentation of farmland target areas is constructed. The land cover classification model performs farmland target area identification and segmentation processing on the satellite image data, and obtains several polygonal farmland areas in the satellite image. The polygonal farmland areas to be studied are then selected.

4. The crop growth monitoring method based on drone inspection route planning and aerial photography data according to claim 1, characterized in that: The method for obtaining a one-way flight inspection route using the UAV inspection planning model includes: S21. Obtain the polygon vertex set of the polygon farmland region boundary and perform centering and covariance matrix calculation, then obtain the eigenvalues ​​of the covariance matrix. With the corresponding first feature vector Second eigenvector ; with the first eigenvector The main direction, with the second feature vector A secondary direction perpendicular to the primary direction; S22. Select the vertex with the longest vertical distance in the main direction as the starting point, and generate several parallel lines parallel to the main direction along the secondary direction according to the flight path spacing; obtain the intersection points of all parallel lines with the boundary of the polygonal farmland area. S23. Using parallel lines as the route segments, starting from the starting point, the parallel lines are sequentially numbered according to their secondary directions; starting from the starting point, each intersection point is sequentially connected by a one-way flight route according to the parallel line number to obtain a set of one-way flight route paths. All route path sets constitute a one-way flight inspection route.

5. The method for monitoring crop growth based on aerial photography data from UAV inspection route planning according to claim 1, characterized in that: In method S2, the drone is equipped with a camera, and the sampling parameters include the ground sampling distance of the images captured by the camera, the image overlap rate, and the shooting tilt angle; the drone's flight altitude. Calculated using the following formula: ,in To convert the tilt angle of the shot into a vertical downward shooting distance, For pixel width, For the width of the camera sensor, The focal length of the camera; Distance between drone flight paths Calculated using the following formula: , This represents the image overlap rate.

6. The crop growth monitoring method based on drone inspection route planning and aerial photography data according to claim 4, characterized in that: In method S21, several sampling points are uniformly interpolated on each edge of the polygon vertex set to form an interpolation point set containing vertices and sampling points. The mean of the interpolation point set is then centered as the centroid. Then, a description is constructed. , The eigenvalues ​​are obtained by eigenvalue decomposition of the 2×2 covariance matrix of directional correlation. With the corresponding first feature vector Second eigenvector .

7. The crop growth monitoring method based on drone inspection route planning and aerial photography data according to claim 4, characterized in that: In method S22, the neighboring edge vectors associated with the starting point of the selected polygonal farmland region are extracted. If the neighboring edge is related to the first feature vector... If the included angle between them is greater than or equal to Q1°, then the principal direction is changed to the adjacent edge vector, and the secondary direction is changed to the direction perpendicular to the principal direction; if the adjacent edge vector and the first eigenvector are... If the included angle is less than Q1°, no changes or adjustments will be made.

8. The method for monitoring crop growth based on aerial photography data of UAV inspection route planning according to claim 4 or 7, characterized in that: In method S22, a latitude and longitude coordinate system is constructed in the satellite map, and polygonal farmland areas, starting points, parallel lines, and intersections are represented in the latitude and longitude coordinate system. In method S23, all path segments are collected in the latitude and longitude coordinate system. Starting from the starting point, the path segments are sequentially numbered according to their secondary directions. The path segment with the highest ranking is selected, and the intersection point closest to the starting point is extracted as the second point of the flight path. The starting point is connected to the second point, and another intersection point of the path segment with the highest ranking is extracted as the third point of the flight path. The intersection point closest to the third point of the path segment with the second ranking is selected as the fourth point of the flight path. The third point and the fourth point of the flight path are connected, and so on, sequentially connecting each intersection point according to the path segment number.

9. The method for monitoring crop growth based on aerial photography data from UAV inspection route planning according to claim 4, characterized in that: In method S22, the parallel lines are all parallel lines that intersect the boundary of the polygonal farmland area, and parallel lines that exceed the boundary of the polygonal farmland area are deleted.

10. The method for monitoring crop growth based on aerial photography data of UAV inspection route planning according to claim 4, characterized in that: In method S23, the connecting paths and all vertices between adjacent path segments are filtered. If the distance between a vertex and the connecting path is greater than the route spacing, the connecting path is split into two segmented paths with that vertex as the path point. If the distance between a vertex and the connecting path is less than or equal to the route spacing, no processing is performed.

Images