Corn leaf crimpness calculation method and system based on image processing

Abstract

The invention discloses a corn leaf crimpness calculation method and system based on image processing. The method comprises the following steps: acquiring image data of corn leaves at multiple positions and fixed points; according to the image data, multi-view corn leaf features are extracted, and leaf three-dimensional image reconstruction of point cloud is set; on the basis of a minimum directed bounding box algorithm, overlapping the three-dimensional image of the blade with a minimum effective bounding box to obtain first geometric data; extracting the first geometric data, calculating the corn leaf crimpness and projection area ratio, and judging the leaf health state; based on the corn leaf crimpness, a curvature thermodynamic diagram is constructed, and based on a time sequence, a dynamic crimping process model is generated, a deformation rate is captured, and a dynamic model is generated, so that an effective method from data acquisition to accurate intervention is provided for intelligent agriculture.

Description

Technical Field

[0001] This invention relates to the field of image processing and recognition technology, and in particular to a method and system for calculating the curvature of corn leaves based on image processing. Background Technology

[0002] Leaf curl is a sensitive indicator of water stress, nutrient deficiency, or pests and diseases in maize. Quantitatively measuring the degree of leaf curl can accurately assess the impact mechanism on the plant and provide effective data for studying the drought resistance and salt tolerance of plants.

[0003] A Chinese invention with application number 202410172883.8 discloses a method, system, device, and medium for calculating the degree of blade curling. The specific method of this invention mainly includes: acquiring a blade image and constructing a triangularly patterned blade model; obtaining multiple grids of the blade model and calculating the vertical vector of each grid, ensuring that the vertical vector length of each grid is the same; obtaining the vertical vector at the centroid position, ensuring that the length of the vertical vector at the centroid position is a multiple of the length of the vertical vector of each grid; projecting the vertical vector of each grid onto the vertical vector at the centroid position and recording the projected length of the vertical vector of each grid, determining the positive or negative zero value of the projected length based on the angle between the direction of the vertical vector of each grid and the direction of the vertical vector at the centroid position; summing the projected lengths, and calculating the blade curling index based on the ratio of the summation result to the length of the vertical vector at the centroid position, thereby determining the blade curling degree.

[0004] The inventions described above fail to combine three-dimensional geometric features with biomechanics, lack the ability to dynamically capture leaf curling, and lack the ability to determine the health status of leaves based on the degree of leaf curling. Summary of the Invention

[0005] The technical problem solved by this invention is that it fails to combine three-dimensional geometric features with biomechanics, lacks the ability to dynamically capture leaf curling, and lacks the ability to determine the health status of leaves based on the degree of leaf curling.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] Firstly, a method for calculating the curl of corn leaves based on image processing includes:

[0008] Step S1: Collect image data of corn leaves at multiple fixed locations;

[0009] Step S2: Based on the image data, extract the features of corn leaves from multiple perspectives and set up a three-dimensional image reconstruction of the leaves from the point cloud.

[0010] Step S3: Based on the minimum directed bounding box algorithm, the three-dimensional image of the blade is overlaid with the minimum effective bounding box to obtain the first geometric data;

[0011] Step S4: Extract the first geometric data, calculate the curling degree and projected area ratio of the corn leaves, and determine the health status of the leaves;

[0012] Step S5: Construct a curvature heatmap based on the curvature of corn leaves;

[0013] Step S6: Generate a dynamic curling process model based on the time series.

[0014] Preferably, step S1 specifically includes:

[0015] Step S11: Equip several industrial cameras and set them in a fixed angled ring around the corn leaves;

[0016] Step S12: Adjust the distance between the camera and the corn leaf according to the size of the corn leaf, and set the industrial camera to high resolution;

[0017] Step S13: Use the controller to set the camera rotation time and image acquisition speed. Set the controller to control several industrial cameras to rotate one revolution per unit time, and automatically take a picture of a corn leaf at regular intervals.

[0018] Step S14: Upload the corn leaf photos collected within a unit of time to the computer and save them as image data.

[0019] Preferably, step S2 specifically includes:

[0020] Step S21: Use the SIFT scale-invariant feature transform algorithm to extract local features from the image data and detect key feature points on the leaf surface. In the unit sample image of the image data, mark the positions of the key feature points, including leaf vein intersections and edge curvature extreme points.

[0021] Step S22: Using Euclidean distance-assisted technology, point cloud background segmentation and point cloud outlier removal are performed on the image data to preserve complete leaf edge information;

[0022] Step S23: Randomly extract two sample images from the image data, image data I, (I∈1,2,3,…,N). Using the position of the leaf vein intersection point, overlap the marked key feature points in the two sample images to obtain a high-dimensional three-dimensional image reconstruction of the leaf.

[0023] Preferably, the specific method for extracting local features from the image data using the SIFT scale-invariant feature transform algorithm includes:

[0024] A Gaussian pyramid is constructed based on the image data through Gaussian smoothing and downsampling. Each sample image I of the image data I(x,y) is downsampled multiple times and a low-resolution layer is generated using Gaussian blurring to form a multi-scale high-dimensional spatial image of the leaf.

[0025] Calculate the difference between the current layer and the upsampled image to obtain the Laplacian layer of the current layer. Based on the multi-scale edge information provided by the Laplacian layer, obtain the key feature points.

[0026] Preferably, the method for using Euclidean distance-assisted technology to perform point cloud background segmentation and point cloud outlier removal on the image data while preserving complete leaf edge information includes:

[0027] Background segmentation is performed based on the key feature nodes. A node P is randomly extracted, and the distance S from the remaining nodes to node P is calculated. The k points closest to P are found as neighboring points. Among the neighboring points, the points whose distance is less than a set threshold are found and used as the leaf index vector clustering point set Q of the point cloud. This process continues until the number of nodes in set Q no longer increases. The nodes in set Q are then used as the edge nodes of the point cloud to complete the background segmentation.

[0028] Calculate the average μ and standard deviation σ of the distances from each node to its K nearest neighbors in set Q. When the average of the K nearest neighbors of a random node in set Q is within the range of μ ± σ × a set threshold, it is determined to be a neighboring point and the node is retained.

[0029] When the average of the K nearest neighbors of a random node in set Q is not within the range of μ±σ×std, it is identified as an outlier and the node is removed.

[0030] Preferably, step S3 specifically includes:

[0031] Key feature nodes from the 3D image of the leaf are projected onto a 3D model (x, y, z), where the z-axis represents the length of the leaf, the x-axis represents the width of the leaf, and the z-axis represents the thickness of the leaf when placed vertically. The minimum and maximum values ​​of the projected nodes in the 3D model are recorded in all three dimensions, resulting in six extreme points. These six extreme points include the minimum leaf width x. min Maximum blade width v max Minimum leaf length z min Maximum blade length z max Minimum y of blade three-dimensional thickness min Maximum y of the blade's three-dimensional thickness max The length L, width W, and height H of the minimum bounding box of the blade are calculated based on the six extreme points. The expression for calculating the length, width, and height of the minimum bounding box is as follows:

[0032] L = z max -z min;

[0033] w = x max -x min ;

[0034] H = y max -y min ;

[0035] For each sample leaf of the three-dimensional leaf image, extract the length, width, height, and volume of the minimum bounding box according to the calculation method of the minimum bounding box, and save them as the first geometric data.

[0036] Preferably, step S4 specifically includes:

[0037] Establish a triangular mesh based on the positions of the leaf point cloud, divide the triangular mesh space into cubic grids of 0.1 cubic millimeters, and overlap the minimum bounding box with the triangular mesh according to the leaf point cloud state;

[0038] Use the convex Hull algorithm to calculate the convex hull volume V of the leaf point cloud leaf , extract the volume V of the minimum bounding box of the first geometric data OBB , calculate the leaf curl C based on the minimum oriented bounding box OBB , the calculation expression of the leaf curl is:

[0039] <000​​​​​​​​​​​​​​​

[0045] When H = 0, set the leaf morphological feature label as the flat area and render it green;

[0046] When 0 < H < the maximum curling threshold, set the leaf morphological feature label as slightly convex and render it yellow;

[0047] When H > the maximum curling threshold, set the leaf morphological feature label as significantly convex and render it red.

[0048] Preferably, step S6 specifically includes:

[0049] For the corn leaves of the same sample, register them according to the time series, obtain the change rate of the key feature nodes of the leaves at adjacent time points, calculate the displacement speed of the leaf vertices, and based on the change rate and displacement speed, dynamically display the visualization image of the curling evolution and generate a dynamic curling process model.

[0050] In a second aspect, a network security protection system based on big data includes an image acquisition module, a feature extraction module, a three-dimensional overlap module, a state detection module, and a visualization module:

[0051] The image acquisition module is used to collect image data of corn leaves at multiple fixed points;

[0052] The feature extraction module is used to extract the features of corn leaves from multiple perspectives according to the image data and set up the three-dimensional image reconstruction of the leaves of the point cloud;

[0053] The three-dimensional overlap module is used to overlap the three-dimensional leaf image with the minimum effective bounding box based on the minimum oriented bounding box algorithm to obtain the first geometric data;

[0054] The state detection module is used to extract the first geometric data, calculate the curling degree and the projection area ratio of the corn leaves, and judge the health state of the leaves;

[0055] The visualization module is used to construct a curvature heat map based on the curling degree of the corn leaves and generate a dynamic curling process model by using the curling process of the leaves recorded under the time series.

[0056] Advantages of the present invention: Through the multi-perspective three-dimensional reconstruction and dynamic modeling technology, the present invention realizes the intelligent and accurate monitoring of the health state of corn leaves. It mainly adopts multi-position fixed-point collection combined with the minimum oriented bounding box algorithm, which can quantify key geometric parameters such as the leaf volume compression rate and the projection area ratio, breaking through the limitations of traditional two-dimensional measurement; innovatively constructs a curvature heat map and a composite curling index to improve the discrimination of leaves; based on the time-series visualization monitoring of leaves, it can capture the shape change rate and generate a dynamic model, providing an effective method from data collection to precise intervention for smart agriculture. Description of the Drawings

[0057] Figure 1 This is a schematic diagram of the basic process of a method for calculating the curvature of corn leaves based on image processing, provided as an embodiment of the present invention. Detailed Implementation

[0058] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0059] Example 1, referring to Figure 1 As one embodiment of the present invention, a method is provided, comprising:

[0060] Step S1: Collect image data of corn leaves at multiple fixed locations;

[0061] Step S2: Based on the image data, extract the features of corn leaves from multiple perspectives and set up the three-dimensional image reconstruction of the leaves from the point cloud.

[0062] Step S3: Based on the minimum directed bounding box algorithm, the three-dimensional image of the blade is overlaid with the minimum effective bounding box to obtain the first geometric data;

[0063] Step S4: Extract the first geometric data, calculate the curling degree and projected area ratio of the corn leaves, and determine the health status of the leaves;

[0064] Step S5: Construct a curvature heatmap based on the curvature of corn leaves;

[0065] Step S6: Generate a dynamic curling process model based on the time series.

[0066] In this embodiment, a minimum oriented bounding box is generated by multi-view 3D reconstruction of corn leaf images, and the geometric data of the leaves is extracted. The 3D projected area and leaf curling degree are calculated based on the geometric data, which effectively solves the problem of insufficient accuracy in corn leaf health monitoring by traditional methods.

[0067] Step S1 specifically includes:

[0068] Step S11: Equip several industrial cameras and set them in a fixed angled ring around the corn leaves;

[0069] Step S12: Adjust the distance between the camera and the corn leaf according to the size of the corn leaf, and set the industrial camera to high resolution;

[0070] Step S13: Use the controller to set the camera rotation time and image acquisition speed. Set the controller to control several industrial cameras to rotate one revolution per unit time, and automatically take a picture of a corn leaf at regular intervals.

[0071] Step S14: Upload the corn leaf photos collected within a unit of time to the computer and save them as image data.

[0072] In this embodiment, a multi-location fixed-point image acquisition method can achieve 360-degree acquisition without blind spots. Through precise timing control, millisecond-level synchronous shooting can be achieved, effectively avoiding image blurring caused by blade shaking. The images are then uploaded to form a standardized image database, providing a high-quality data foundation for subsequent analysis.

[0073] Step S2 specifically includes:

[0074] Step S21: Use the SIFT scale-invariant feature transform algorithm to extract local features from the image data and detect key feature points on the leaf surface. In the unit sample image of the image data, the key feature points are marked with their positions. Key feature points include leaf vein intersections and edge curvature extreme points.

[0075] Step S22: Using Euclidean distance-assisted technology, point cloud background segmentation and point cloud outlier removal are performed on the image data to preserve complete leaf edge information;

[0076] Step S23: Randomly extract two sample images from the image data, image data I, (I∈1,2,3,…,N). Using the position of the leaf vein intersection point, overlap the marked key feature points in the two sample images to obtain a high-dimensional three-dimensional image reconstruction of the leaf.

[0077] In this embodiment, the accuracy and efficiency of 3D modeling are significantly improved by using feature extraction box point cloud processing technology. The SIFT algorithm is used to accurately locate key features such as leaf vein intersections and edge curvature extreme points, thereby improving the accuracy of feature point detection. Based on Euclidean distance-assisted technology, intelligent point cloud denoising is achieved, which removes background interference points while preserving complete leaf edge information, improving the reconstruction environment and providing a high-fidelity 3D data foundation for subsequent curvature analysis and health diagnosis.

[0078] Specific methods for extracting local features from image data using the SIFT scale-invariant feature transform algorithm include:

[0079] Gaussian pyramids are constructed based on image data through Gaussian smoothing and downsampling. Each sample image I of image data I(x,y) is downsampled multiple times and Gaussian blur is used to generate a low-resolution layer, forming a multi-scale high-dimensional spatial image of the leaf.

[0080] Calculate the difference between the current layer and the upsampled image to obtain the Laplacian layer of the current layer. Based on the multi-scale edge information provided by the Laplacian layer, obtain the key feature points.

[0081] In this embodiment, the multi-scale feature extraction method of SIFT algorithm obtains multi-scale edge information through innovative Gaussian pyramid construction and Laplacian difference processing, providing high-matching features for 3D reconstruction and significantly improving the accuracy and robustness of maize leaf feature detection.

[0082] Methods for performing point cloud background segmentation and outlier removal on image data using Euclidean distance-assisted techniques while preserving complete leaf edge information include:

[0083] Background segmentation is performed based on key feature nodes. A node P is randomly extracted, and the distance S from the remaining nodes to node P is calculated. The k points closest to P are found as neighboring points. Among the neighboring points, the points whose distance is less than a set threshold are found as the leaf index vector clustering point set Q of the point cloud. This process continues until the number of nodes in set Q no longer increases. The nodes in set Q are then used as the edge nodes of the point cloud to complete the background segmentation.

[0084] Calculate the average μ and standard deviation σ of the distances from each node to its K nearest neighbors in set Q. When the average of the K nearest neighbors of a random node in set Q is within the range of μ ± σ × a set threshold, it is determined to be a neighboring point and the node is retained.

[0085] When the average of the K nearest neighbors of a random node in set Q is not within the range of μ±σ×std, it is identified as an outlier and the node is removed.

[0086] In this embodiment, the dynamic distance threshold is set to 3 times the average spacing of the point cloud. An adaptive region growing algorithm based on feature nodes is adopted to achieve accurate separation of the background and leaves through the dynamic distance threshold. Combined with K-nearest neighbor statistical analysis and the 3-times standard deviation principle, the distance distribution of the local neighborhood is calculated to effectively identify and remove noise nodes.

[0087] Step S3 specifically includes:

[0088] Key feature nodes from the 3D image of the blade are projected onto a 3D model (x, y, z). The z-axis of the 3D model represents the length of the blade, the x-axis represents the width of the blade, and the z-axis represents the thickness of the blade when placed vertically. The minimum and maximum values ​​of the projected nodes in the 3D model are recorded in the three dimensions, obtaining six extreme points. These six extreme points include the minimum blade width x. min Maximum blade width x max Minimum leaf length z min Maximum blade length z max Minimum y of blade three-dimensional thickness min Maximum y of the blade's three-dimensional thickness max The length L, width W, and height H of the minimum bounding box of the blade are calculated based on the six extreme points. The expression for calculating the length, width, and height of the minimum bounding box is as follows:

[0089] L = z max -z min ;

[0090] w = x max -x min ;

[0091] H = y max -y min ;

[0092] For each sample leaf in the 3D image of the leaf, the length, width, height and volume of the minimum bounding box are extracted according to the minimum bounding box calculation method and saved as the first geometric data.

[0093] In this embodiment, a dedicated three-dimensional coordinate system for the blade is established. The point cloud feature nodes are accurately projected onto the (x,y,z) space along the main vein direction with the Z-axis. An extreme point capture algorithm is used to obtain boundary values ​​in six dimensions to control the measurement error of the bounding box size. The complex three-dimensional morphology of the blade is abstracted into four core parameters: length, width, height, and volume, which improves the efficiency of batch processing. In addition, an extreme point iterative correction algorithm is used to automatically adapt to blades in different curling states.

[0094] Step S4 specifically includes:

[0095] A triangular mesh is created based on the position of the blade point cloud. The space of the triangular mesh is divided into a cubic mesh of 0.1 cubic millimeters. The minimum bounding box is overlapped with the triangular mesh according to the state of the blade point cloud.

[0096] The convex hull volume V of the leaf point cloud is calculated using the convex Hull algorithm. leaf Extract the volume V of the minimum bounding box of the first geometric data. OBB Calculate the blade curl C based on the minimum directed bounding box. OBB The formula for calculating blade curl is:

[0097]

[0098] The Teasar algorithm is used to extract the center line of the leaf veins, and the midrib direction vector angle is obtained by linear fitting using the least squares method. Parallel rays are projected according to the shortest axis direction of the minimum bounding box, and the intersection points of the projected parallel rays and the triangular mesh are obtained. The effective area of ​​all leaves hit by the rays on the triangular mesh is counted. The effective area includes the actual surface area and the projected area. The area ratio of the actual surface area to the projected area is calculated. Based on the health status assessment matrix, the leaf curl, area ratio, and midrib direction vector angle are distinguished according to the set threshold range to obtain the health status of the corn leaves.

[0099] In this embodiment, a health status assessment matrix is ​​set:

[0100] When the leaf curl is less than 0.3, the leaf is in a healthy state. When the leaf curl is between 0.3 and 0.6, the leaf is in a mild stress state. When the leaf curl is greater than 0.6, the leaf is in a severe stress state;

[0101] When the area ratio is between 1.0 and 1.2, the leaf is in a healthy state. When the area ratio is between 1.2 and 1.8, the leaf is in a mild stress state. When the area ratio is greater than 1.8, the leaf is in a severe stress state;

[0102] When the main vein direction vector angle is less than 12°, the leaf is in a healthy state. When the main vein direction vector angle is between 15° and 30°, the leaf is in a mild stress state. When the main vein direction vector angle is greater than 30°, the leaf is in a severe stress state;

[0103] The leaf curl is calculated by the volume of the convex hull and the volume of the minimum oriented bounding box. Three key features, namely the spatial compression ratio, surface area ratio, and main vein offset angle, are integrated. And through the donation status evaluation matrix, the health status of corn leaves is distinguished. An optical correction technology is also used. Based on the parallel light projection algorithm in the shortest axis direction, the projected area of the intersection of the light and the grid is calculated to automatically compensate for the projection deviation caused by the difference in leaf thickness.

[0104] Step S5 specifically includes:

[0105] According to the three-dimensional triangular mesh and leaf curl, the average curvature H is obtained by quadratic surface fitting, and the outliers are removed by curvature filtering. The average curvature is used to distinguish the morphological features according to the curvature setting range, and the corresponding colors are rendered:

[0106] H < the lowest curl threshold, set the leaf morphological feature label as severely concave, and render dark blue;

[0107] The lowest curl threshold < H < 0, set the leaf morphological feature label as mildly concave, and render light blue;

[0108] H = 0, set the leaf morphological feature label as flat area, and render green;

[0109] 0 < H < the highest curl threshold, set the leaf morphological feature label as mildly convex, and render yellow;

[0110] H > the highest curl threshold, set the leaf morphological feature label as significantly convex, and render red.

[0111] In this embodiment, through the three-dimensional morphological visualization analysis method of corn leaves based on curvature features, multi-scale curvature quantization and intelligent rendering technology, the intuitive and accurate diagnosis of crop physiological status is achieved.

[0112] Step S6 specifically includes:

[0113] Based on the same sample of maize leaves, the rate of change of key feature nodes of leaves at adjacent time points is obtained by time series registration, and the displacement velocity of the leaf apex is calculated. Based on the rate of change and displacement velocity, a dynamic visualization image of the curling evolution is displayed, and a dynamic curling process model is generated.

[0114] In this embodiment, the dynamic curling process modeling method is used to reconstruct crop growth from the time-series images, thereby realizing the dynamic monitoring of the corn leaf growth process, constructing a spatiotemporal fusion model, dynamically displaying the 24-hour movement trajectory of feature points, and automatically acquiring the volume change curve, so that the complex physiological process can be presented intuitively.

[0115] Example 2 is another embodiment of the present invention. Unlike the first embodiment, this embodiment provides a calculation system for the curling degree of corn leaves based on image processing. To verify and illustrate the technical effect of the method, this embodiment compares the traditional technical solution with the method of the present invention, and compares the experimental results with scientific demonstration methods to verify the real effect of the method.

[0116] A system for calculating the curl of corn leaves based on image processing includes: an image acquisition module, a feature extraction module, a three-dimensional overlap module, a state detection module, and a visualization module;

[0117] The image acquisition module is used to acquire image data of corn leaves at multiple fixed locations.

[0118] The feature extraction module is used to extract multi-view corn leaf features based on image data and set up point cloud-based 3D image reconstruction of the leaves.

[0119] The 3D overlay module is used to overlay the 3D image of the blade with the minimum effective bounding box based on the minimum directed bounding box algorithm to obtain the first geometric data.

[0120] The status detection module is used to extract the first geometric data, calculate the curling degree and projected area ratio of corn leaves, and determine the health status of the leaves;

[0121] The visualization module is used to construct a curvature heatmap based on the curvature of corn leaves and to generate a dynamic curvature process model using the leaf curvature process recorded in the time series.

[0122] In this embodiment, the invention is divided into five modules through module description, which comprehensively monitor the curling degree of corn leaves. Based on the superposition and fusion of three-dimensional key points and minimum directed bounding boxes of multi-module collaborative work, the dynamic leaf curling degree is calculated. Based on the analysis of data such as curling degree and volume, the crop growth status is quantitatively evaluated, and the crop growth status is efficiently monitored.

[0123] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0124] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for calculating the curl degree of corn leaves based on image processing, characterized in that, include: Step S1: Collect image data of corn leaves at multiple fixed locations; Step S2: Based on the image data, extract the features of corn leaves from multiple perspectives and set up a three-dimensional image reconstruction of the leaves from the point cloud. Step S3: Based on the minimum directed bounding box algorithm, the three-dimensional image of the blade is overlaid with the minimum effective bounding box to obtain the first geometric data; Step S4: Extract the first geometric data, calculate the curling degree and projected area ratio of the corn leaves, and determine the health status of the leaves; Step S5: Construct a curvature heatmap based on the curvature of corn leaves; Step S6: Generate a dynamic curling process model based on the time series.

2. The method for calculating the curl of corn leaves based on image processing as described in claim 1, characterized in that: Step S1 specifically includes: Step S11: Equip several industrial cameras and set them in a fixed angled ring around the corn leaves; Step S12: Adjust the distance between the camera and the corn leaf according to the size of the corn leaf, and set the industrial camera to high resolution; Step S13: Use the controller to set the camera rotation time and image acquisition speed. Set the controller to control several industrial cameras to rotate one revolution per unit time, and automatically take a picture of a corn leaf at regular intervals. Step S14: Upload the corn leaf photos collected within a unit of time to the computer and save them as image data.

3. The method for calculating the curl of corn leaves based on image processing as described in claim 2, characterized in that: Step S2 specifically includes: Step S21: Use the SIFT scale-invariant feature transform algorithm to extract local features from the image data and detect key feature points on the leaf surface. In the unit sample image of the image data, mark the positions of the key feature points, including leaf vein intersections and edge curvature extreme points. Step S22: Using Euclidean distance-assisted technology, point cloud background segmentation and point cloud outlier removal are performed on the image data to preserve complete leaf edge information; Step S23: Randomly extract two sample images from the image data, image data I, (I∈1,2,3,…,N). Using the position of the leaf vein intersection point, overlap the marked key feature points in the two sample images to obtain a high-dimensional three-dimensional image reconstruction of the leaf.

4. The method for calculating the curl of corn leaves based on image processing as described in claim 3, characterized in that: The specific method for extracting local features from the image data using the SIFT scale-invariant feature transform algorithm includes: A Gaussian pyramid is constructed based on the image data through Gaussian smoothing and downsampling. Each sample image I of the image data I(x,y) is downsampled multiple times and a low-resolution layer is generated using Gaussian blurring to form a multi-scale high-dimensional spatial image of the leaf. Calculate the difference between the current layer and the upsampled image to obtain the Laplacian layer of the current layer. Based on the multi-scale edge information provided by the Laplacian layer, obtain the key feature points.

5. The method for calculating the curl of corn leaves based on image processing as described in claim 4, characterized in that: The method for performing point cloud background segmentation and outlier removal on the image data using Euclidean distance-assisted technology while preserving complete leaf edge information includes: Segment the background based on the key feature nodes. Randomly extract a node P, calculate the distance S from the remaining nodes to node P, find k points closest to point P as neighboring points, and find the points among the neighboring points with a distance less than the set threshold as the clustering point set Q of the leaf index vector of the point cloud. Stop when the number of nodes in set Q no longer increases. Take the nodes in set Q as the edge nodes of the point cloud to complete the background segmentation; Calculate the average value μ and standard deviation σ of the distances from each node in set Q to its K closest distance points. When the average value of the K nearest points of a random node in set Q is within the range of μ ± σ × set threshold, it is determined as a neighborhood point and the node is retained; When the average value of the K nearest points of a random node in set Q is not within the range of μ ± σ × std, it is determined as an outlier and the node is removed.

6. The method for calculating the curl of corn leaves based on image processing as described in claim 5, characterized in that: Step S3 specifically includes: Key feature nodes from the 3D image of the leaf are projected onto a 3D model (x, y, z), where the z-axis represents the length of the leaf, the x-axis represents the width of the leaf, and the z-axis represents the thickness of the leaf when placed vertically. The minimum and maximum values ​​of the projected nodes in the 3D model are recorded in all three dimensions, resulting in six extreme points. These six extreme points include the minimum leaf width x. min Maximum blade width x max Minimum leaf length z min Maximum blade length z max Minimum y of blade three-dimensional thickness min Maximum y of the blade's three-dimensional thickness max The length L, width W, and height H of the minimum bounding box of the blade are calculated based on the six extreme points. The expression for calculating the length, width, and height of the minimum bounding box is as follows: L=z max -With min ; w=x max -x min ; H=y max -y min ; For each sample leaf of the three-dimensional leaf image, extract the length, width, height, and volume of the minimum bounding box according to the calculation method of the minimum bounding box, and save them as the first geometric data.

7. The method for calculating the curl of corn leaves based on image processing as described in claim 6, characterized in that: Step S4 specifically includes: Establish a triangular mesh based on the position of the leaf point cloud, divide the triangular mesh space into cubic grids of 0.1 cubic millimeters, and overlap the minimum bounding box with the triangular mesh according to the state of the leaf point cloud; The convex hull volume V of the leaf point cloud is calculated using the convex Hull algorithm. leaf Extract the volume V of the minimum bounding box of the first geometric data. OBB Calculate the blade curl C based on the minimum directed bounding box. OBB The formula for calculating the blade curl is: Use the Teasar algorithm to extract the centerline of the leaf veins, obtain the main vein direction vector angle through linear fitting by the least squares method, project parallel light rays according to the shortest axis direction of the minimum bounding box, obtain the intersection points of the projected parallel light rays and the triangular mesh, count the sum of the effective areas of the projections of all leaves hit by the light rays on the triangular mesh. The sum of the effective areas includes the actual surface area and the projected area. Calculate the area ratio of the actual surface area and the projected area. According to the health status evaluation matrix, distinguish the leaf curl, area ratio, and main vein direction vector angle within the set threshold range to obtain the health status of the corn leaf.

8. The method for calculating the curl of corn leaves based on image processing as described in claim 7, characterized in that: Step S5 specifically includes: According to the three-dimensional triangular mesh and the leaf curl, obtain the average curvature H through quadratic surface fitting, and use curvature filtering to remove outliers. Distinguish the morphological features of the average curvature within the set curvature range and render the corresponding colors: H < the lowest curl threshold, set the leaf morphological feature label as severely concave, and render dark blue; The lowest curl threshold < H < 0, set the leaf morphological feature label as mildly concave, and render light blue; H = 0, set the leaf morphological feature label as flat area, and render green; 0 < H < the highest curl threshold, set the leaf morphological feature label as mildly convex, and render yellow; H > the highest curl threshold, set the leaf morphological feature label as significantly convex, and render red.

9. The method for calculating the curl of corn leaves based on image processing as described in claim 8, characterized in that: Step S6 specifically includes: According to the corn leaves of the same sample, perform registration according to the time series, obtain the change rate of the key feature nodes of the leaves at adjacent time points, and calculate the displacement speed of the leaf vertices. According to the change rate and displacement speed, dynamically display the visualization image of the curl evolution and generate a dynamic curl process model.

10. A system for calculating the curvature of corn leaves based on image processing, implemented according to any one of claims 1-9, characterized in that: It includes a picture acquisition module, a feature extraction module, a three-dimensional overlap module, a state detection module, and a visualization module: The image acquisition module is used to acquire image data of corn leaves at multiple fixed locations; The feature extraction module is used to extract multi-view corn leaf features based on the image data and set up point cloud-based three-dimensional image reconstruction of the leaves. The three-dimensional overlap module is used to overlap the three-dimensional image of the blade with the minimum effective bounding box based on the minimum directed bounding box algorithm to obtain the first geometric data. The state detection module is used to extract the first geometric data, calculate the curling degree and projected area ratio of the corn leaves, and determine the health status of the leaves; The visualization module is used to construct a curvature heatmap based on the curvature of corn leaves, and to generate a dynamic curvature process model using the leaf curvature process recorded in the time series.

Images