Plant phenotype monitoring method based on computer vision

Abstract

The invention relates to the technical field of plant phenotypes, and discloses a plant phenotype monitoring method based on computer vision, which comprises the following steps: setting a growth cycle, collecting data of potted plants which have the same growth cycle and grow normally, obtaining a top view and a side view of the plants, determining convex hull nodes, forming convex hulls, and collecting canopy height data. Leaf breakage is recognized by segmenting a convex hull, calculating an edge difference value, detecting whether the canopy height is abnormal or not and executing a slope judgment strategy. And if the plant grows normally, executing a plant centrifugation strategy, judging the growth state of the plant through comparison of density areas, and finally marking that the plant grows normally or abnormally. The method improves the accuracy of plant phenotype monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of plant phenotype, and in particular to a plant phenotype monitoring method based on computer vision. Background Art

[0002] Plant phenomics is the science of studying plant phenotypes, aiming to collect and analyze large-scale, accurate phenotypic data of plants through high-throughput phenotyping technology. Traditional plant phenotyping monitoring methods mainly rely on manual measurement, which is inefficient, costly, and highly subjective, and cannot meet the needs of modern agriculture and horticulture for high-throughput, accurate plant phenotyping monitoring. With the rapid development of computer vision technology, computer vision-based plant phenotyping monitoring methods have gradually become a research hotspot and have shown great application potential.

[0003] Traditional plant phenotyping methods mainly include manual measurement and image analysis. Manual measurement methods rely on manual visual inspection and manual measurement, such as measuring plant height, leaf area, biomass and other indicators. This method is time-consuming, costly, and easily affected by human factors, making it difficult to achieve large-scale, high-throughput monitoring. Image analysis methods use image processing technology to analyze plant images and extract the morphological characteristics of plants. Early image analysis methods were mainly based on artificially designed features, such as color, texture, etc., with limited accuracy and difficulty in processing complex image data.

[0004] In recent years, with the advancement of computer vision technology, plant phenotyping monitoring methods based on computer vision have made significant progress. These methods use computer vision technology to automatically analyze plant images, extract plant morphological characteristics, and evaluate plant growth status based on these characteristics.

[0005] Although plant phenotypic monitoring technology based on computer vision has made significant progress, there are still some shortcomings, such as limited recognition accuracy of plant morphological changes and lack of comprehensive analysis of plant growth abnormalities. The present invention proposes a multi-dimensional, multi-stage plant phenotypic monitoring method based on computer vision, aiming to solve the above problems. The present invention combines multiple technical means such as canopy height measurement, convex hull contour analysis, leaf density analysis, and leaf slope analysis to conduct a multi-dimensional comprehensive evaluation of plant growth status, thereby effectively improving monitoring accuracy and reliability. In particular, through the analysis of the side view, key information that can better reflect the growth status of the plant is extracted, and through reasonable processing of the data, a more accurate basis for judgment is obtained. In addition, the present invention also proposes a new algorithm to determine the convex hull nodes, which avoids the misjudgment that may occur in the traditional method and improves the accuracy of the judgment.

[0006] The present invention can more effectively monitor the growth status of plants, and timely discover and handle growth abnormalities, providing powerful technical support for precision agriculture and horticulture. Summary of the invention

[0007] The present invention provides a plant phenotype monitoring method based on computer vision, which is used to promote the solution of the problems mentioned in the above background technology.

[0008] The present invention provides the following technical solution: a plant phenotype monitoring method based on computer vision, comprising:

[0009] Obtain all potted plants with the same growth cycle and normal growth status, and enter them into a qualified set;

[0010] Potted plants consist of plants and pots, and the states of potted plants include normal growth and abnormal growth;

[0011] The plant phenotypic data collection strategy is sequentially executed on the elements in the qualified set, and the canopy height set and the reasonable difference set are obtained according to the results of the plant phenotypic data collection strategy execution;

[0012] When all elements in the qualified set have completed the plant phenotype data collection strategy, the potted plants that need to be tested are tested in turn;

[0013] If the detection result shows that the canopy height of the detected potted plant is abnormal, the detection is stopped, otherwise, the convex hull detection strategy is executed;

[0014] If an abnormal leaf vertex is found during the execution of the convex hull detection strategy, the slope judgment strategy is executed on it to determine whether the leaf to which the abnormal leaf vertex belongs is broken;

[0015] Determine whether to execute the plant centrifugal strategy according to the execution result of the convex hull contour detection strategy;

[0016] If the plant centrifugal strategy is executed, the potted plant state is set according to the execution result, otherwise, the detection is stopped.

[0017] Optionally, the plant phenotype data collection strategy is executed on the elements in the qualified set in sequence, and the canopy height set and the reasonable difference set are obtained according to the results of the plant phenotype data collection strategy execution:

[0018] The strategy for collecting plant phenotypic data is as follows:

[0019] Get top and side views of potted plants;

[0020] Among the four edges of the side view, the edge closest to the bottom of the flower pot in the side view is taken as the horizontal edge, and the convex hull nodes are determined according to the horizontal edge and the leaf vertices on the side view;

[0021] Taking the convex hull node determined by the horizontal edge and the leaf node on the side view as the starting point, connect with the remaining leaf nodes in sequence, and record the formed closed graph as the convex hull;

[0022] The point where the bottom of the plant touches the pot in the side view is recorded as the growth point;

[0023] The canopy height is collected for the convex hull according to the growth point, and the collection result is recorded into the canopy height set;

[0024] Draw straight lines through the growth points to segment the convex hull, and collect and process data on the segmented convex hull.

[0025] Optionally, taking one of the four edges of the side view that is closest to the bottom of the flower pot in the side view as a horizontal edge, and determining a convex hull node according to the horizontal edge and a leaf vertex on the side view, includes:

[0026] Get a top view of the potted plant;

[0027] Mark all leaf vertices in the top view;

[0028] The leaf apex is the tip of each leaf in a potted plant;

[0029] Connect the leaf vertices in the top view in pairs to form multiple connection lines;

[0030] Select the longest connecting line as line segment 1;

[0031] In the top view, draw a straight line perpendicular to line segment 1 through the midpoint of line segment 1, which is recorded as the shooting straight line;

[0032] An image of the potted plant is acquired along a shooting straight line and in a direction pointing toward the potted plant, which is recorded as a side view;

[0033] Randomly select an edge perpendicular to the horizontal edge in the side view as the vertical edge;

[0034] Mark all leaf vertices in the side view;

[0035] Obtain the straight-line distance between each leaf vertex and the horizontal edge in the side view respectively, and enter the leaf vertex with the smallest straight-line distance into the vertex set;

[0036] If the number of leaf vertices in the vertex set is 1, the leaf vertices in the vertex set are used as convex hull nodes;

[0037] If the number of leaf vertices in the vertex set is greater than 1, the straight-line distance between each leaf vertex in the vertex set and the vertical edge is obtained, and the leaf vertex corresponding to the minimum straight-line distance is selected as the convex hull node.

[0038] Optionally, the method of taking the convex hull node determined by the horizontal edge and the leaf node on the side view as the starting point, sequentially connecting with the remaining leaf nodes, and recording the formed closed graph as the convex hull includes:

[0039] S1. Draw a straight line passing through the convex hull nodes and parallel to the horizontal edge, recorded as the convex hull straight line;

[0040] S2. If there are leaf vertices other than the convex hull node on the convex hull line, connect the convex hull node and the leaf vertex on the convex hull line that is farthest from the convex hull node to form a convex hull connecting line segment;

[0041] Select the leaf vertex farthest from the convex hull node on the convex hull line as the new convex hull node, and rotate the convex hull line counterclockwise around the convex hull node;

[0042] S3. If the convex hull line only contains convex hull nodes, rotate the convex hull line counterclockwise around the convex hull nodes;

[0043] S4. If the convex hull line passes through the remaining leaf vertices except the convex hull nodes during the rotation process, then perform operations S2 to S4;

[0044] S5. If the convex hull straight line touches the convex hull connecting line segment during the rotation process, the rotation is stopped;

[0045] The figure formed by connecting all the convex hull segments in the side view is recorded as the convex hull.

[0046] Optionally, collecting the canopy heights of the convex hull according to the growth points and recording the collected results into a canopy height set includes:

[0047] Get all leaf vertices on the convex hull edge contour;

[0048] The leaf vertex farthest from the horizontal edge is selected as the canopy point;

[0049] Draw a straight line through the growth point parallel to the horizontal edge, record the vertical distance between the canopy point and the straight line as the canopy height, and enter it into the canopy height set.

[0050] Optionally, the step of dividing the convex hull by drawing straight lines through the growth points, and collecting and processing data on the divided convex hull includes:

[0051] Draw a straight line perpendicular to the horizontal edge through the growth point, which is recorded as the first dividing line;

[0052] The intersection point of the first dividing line and the convex hull is recorded as the first intersection point and the second intersection point;

[0053] The straight-line distance between the first intersection point and the horizontal edge is greater than that between the second intersection point and the horizontal edge;

[0054] Dividing the convex hull into a first convex hull and a second convex hull by a first dividing straight line;

[0055] S6. Draw a straight line through the first intersection point parallel to the horizontal edge, which is recorded as the collection line;

[0056] Set the movement threshold for movement;

[0057] S7, moving the collection line along the first segmentation line toward the horizontal edge, with the moving distance being equal to the moving threshold;

[0058] S8, obtaining the intersection points of the moving collection line with the first convex hull, the second convex hull and the first segmentation line, and recording them as the first convex hull point, the first segmentation point and the second convex hull point respectively;

[0059] The first convex hull point is located at the first convex hull edge contour, and the second convex hull point is located at the second convex hull edge contour;

[0060] S9, calculating the difference between the distance from the first convex hull point to the first segmentation point and the distance from the second convex hull point to the first segmentation point, and recording the difference into an edge difference set;

[0061] Repeating S6 to S9 until the growth point is within the collection line;

[0062] If all elements in the qualified set are traversed, the minimum and maximum values ​​in the edge difference set are selected, and the range formed between the minimum and maximum values ​​is recorded as the reasonable difference interval.

[0063] Optionally, when all elements in the qualified set have completed the plant phenotype data collection strategy, the potted plants that need to be tested are tested in sequence, including:

[0064] Execute a plant phenotypic data collection strategy on the detected potted plants, obtain the canopy height of the detected potted plants according to the collection results, and determine whether the canopy height is within the canopy height set;

[0065] According to the canopy height judgment result, the canopy height status of the detected potted plant is marked to determine whether to execute the convex hull contour detection strategy, as follows:

[0066] Canopy height status includes abnormal height and normal height;

[0067] Get all potted plants that need to be tested and enter them into the test set;

[0068] Randomly select any element in the detection set as the detection element, and obtain the growth point and horizontal edge in the side view of the detection element;

[0069] Draw a straight line perpendicular to the horizontal edge through the growth point, which is recorded as the second dividing line;

[0070] The intersection points of the second segmentation straight line and the edge contour of the detected convex hull are recorded as the third intersection point and the fourth intersection point;

[0071] The straight-line distance between the third intersection point and the horizontal edge is greater than that between the fourth intersection point and the horizontal edge;

[0072] Divide the convex hull into a third convex hull and a fourth convex hull by a second dividing line;

[0073] Get all leaf vertices on the edge contour of the detected convex hull;

[0074] Select the leaf vertex farthest from the horizontal edge as the canopy point;

[0075] Draw a straight line through the growth point parallel to the horizontal edge, and record the vertical distance between the crown point and the straight line as the crown height;

[0076] If the canopy height of the detection element is less than the minimum value in the canopy height set or greater than the maximum value in the canopy height set, the canopy height of the detection element is marked as abnormal height, and any element in the detection set is reselected as a new detection element, and the plant phenotypic data collection strategy is executed;

[0077] If the canopy height of the detected element is greater than or equal to the minimum value in the canopy height set and less than or equal to the maximum value in the canopy height set, the convex hull contour detection strategy is executed.

[0078] Optionally, if the detection result shows that the canopy height of the detected potted plant is abnormal, the detection is stopped; otherwise, the convex hull detection strategy is executed, including:

[0079] The convex hull detection strategy is as follows:

[0080] S10. Draw a straight line through the third intersection point parallel to the horizontal edge, which is recorded as the test line;

[0081] Set the distance threshold used to test the line movement distance;

[0082] S11, moving the test line toward the horizontal edge along the second dividing line, and the moving distance is the distance threshold;

[0083] S12, obtaining the intersection points of the moved test line with the third convex hull, the fourth convex hull and the second segmentation straight line, and recording them as the third convex hull point, the second segmentation point and the fourth convex hull point respectively;

[0084] The third convex hull point is located at the edge contour of the third convex hull, and the fourth convex hull point is located at the edge contour of the fourth convex hull;

[0085] S13, calculating the difference between the distance from the third convex hull point to the second segmentation point and the distance from the fourth convex hull point to the second segmentation point;

[0086] If the above difference is within a reasonable difference range, then execute S14 operation;

[0087] If the difference is not within a reasonable difference range, the third convex hull point and the fourth convex hull point currently corresponding to the difference are marked as abnormal nodes, and operations S11 to S14 are performed;

[0088] S14, in the test line, select the midpoint between the third convex hull point and the second segmentation point, and the midpoint between the fourth convex hull point and the second segmentation point as density nodes;

[0089] The operations S10 to S14 are repeatedly performed until the growth point is on the test line.

[0090] Optionally, if an abnormal leaf vertex is found during the execution of the convex hull detection strategy, a slope judgment strategy is executed on it to judge whether the leaf to which the abnormal leaf vertex belongs is broken, including:

[0091] Obtain all abnormal nodes on the edge contour of the third convex hull, and select the abnormal nodes closest and farthest from the horizontal edge as interception points respectively;

[0092] The contour edge between the interception points on the third convex hull contour edge is recorded as an abnormal line segment, and the slope judgment strategy is executed as follows:

[0093] Get all leaf vertices on the abnormal line segment and enter them into the abnormal set in descending order according to their distance from the horizontal edge;

[0094] Sequentially obtain the leaf vertices in the exception set as the target leaf vertices;

[0095] Get the leaf contour corresponding to the vertex of the target leaf on the side view, and record it as the target contour;

[0096] The target contour includes an upper contour and a lower contour;

[0097] Taking the target leaf vertex as the starting point, set multiple tangent points with equal intervals on the upper contour in sequence;

[0098] Draw tangent lines tangent to the upper contour through the tangent points respectively, and enter the slope corresponding to each tangent line into the slope set in sequence;

[0099] Set the slope threshold used to determine the slope change;

[0100] The differences of adjacent slopes in the slope set are recorded in turn into the slope change set;

[0101] If there is an element in the slope change set that is greater than the slope threshold, the target node is marked as a normal node, and the leaf vertex on the abnormal line segment is reselected as the new target leaf vertex, and the slope judgment strategy is executed;

[0102] If all elements in the slope change set are less than the slope threshold, the current potted plant state is marked as abnormal growth, and the slope judgment strategy is stopped;

[0103] If all leaf vertices in the abnormal line segment of the third convex hull are traversed and the current potted plant state is not marked as abnormal growth, the abnormal line segment of the fourth convex hull is obtained and the slope judgment strategy is executed.

[0104] Optionally, judging whether to execute the plant centrifugal strategy according to the execution result of the convex hull contour detection strategy includes:

[0105] The plant centrifugation strategy is as follows:

[0106] If the slope judgment strategy shows normal growth, the plant centrifugal strategy is implemented for the potted plants, as follows:

[0107] Randomly select a density node in the side view as the starting point;

[0108] S15, starting from the starting point, in a counterclockwise order, select a density node adjacent to the starting point as the next density node;

[0109] S16, connecting the starting point and the next density node selected;

[0110] S17, taking the next selected density node as a new starting point;

[0111] Repeat steps S16 to S17 until all density nodes are connected to form a closed graph, which is recorded as a density region;

[0112] The second dividing line divides the density area into a first density contrast area and a second density contrast area;

[0113] Set the interval threshold for multiple segmentation line intervals;

[0114] Draw multiple straight lines parallel to the horizontal edge, recorded as density segmentation lines, and the interval between adjacent density segmentation lines is equal to the interval threshold;

[0115] The density segmentation straight line divides the first density contrast region and the second density contrast region into a plurality of sub-density contrast regions respectively;

[0116] Acquire all sub-density comparison areas in the first density comparison area;

[0117] Obtaining the leaf contour area of ​​each sub-density contrast region within the first density contrast region by computer vision;

[0118] According to the order of the distance between the sub-density comparison area and the horizontal edge, the leaf contour area in the first density comparison area is recorded into the first area set from far to near;

[0119] Acquire all sub-density comparison areas in the second density comparison area;

[0120] Obtaining the leaf contour area of ​​each sub-density contrast region within the second density contrast region by computer vision;

[0121] According to the order of the distance between the sub-density comparison area and the horizontal edge, the leaf contour area in the second density comparison area is recorded into the second area set from far to near;

[0122] Set the area threshold used to determine the difference in leaf outline area;

[0123] Select the first element in the first area set and the first element in the second area set respectively, and calculate the absolute value of their difference, which is recorded as the area difference;

[0124] If the area difference is less than or equal to the area threshold, then reselect elements in the first area set and the second area set in turn and calculate the area difference;

[0125] If the area difference is greater than the area threshold, the current potted plant status is marked as abnormal growth;

[0126] If all elements in the first area set and the second area set are traversed and the area difference corresponding to each element is less than or equal to the area threshold, the plant growth state of the current detection element is marked as normal growth.

[0127] The present invention has the following beneficial effects:

[0128] 1. This is a computer vision-based plant phenotyping monitoring method. All leaf vertices are marked through a top view, and the longest connecting line is found as line segment No. 1. Then, with the midpoint of this line segment as the center, a side view is obtained perpendicular to the line segment direction. By connecting the leaf vertices in the top view two by two, the most suitable angle can be found to comprehensively collect plant phenotypic data, providing complete data support for subsequent analysis. In the side view, an edge perpendicular to the horizontal edge is randomly selected as the vertical edge, and all leaf vertices are marked. Next, the straight-line distance from each leaf vertex to the horizontal edge is calculated, and the leaf vertices with the smallest distance are selected to form a vertex set. If there is only one leaf vertex in the set, it is directly used as a convex hull node; if there are multiple leaf vertices, the leaf vertex with the smallest distance to the vertical edge is further selected as the convex hull node; by combining the top view and the side view, the vertex position of the plant leaf can be determined more accurately, thereby improving the accuracy of the convex hull node determination; by selecting the leaf vertex with the smallest distance to the horizontal and vertical edges as the convex hull node, it can better adapt to plants of different shapes and sizes, and improve the robustness of the system.

[0129] 2. This is a computer vision-based plant phenotyping monitoring method. By making a convex hull line parallel to the horizontal edge, and finding the farthest leaf vertex except the convex hull node on the line, connecting the two points to form a convex hull connecting line segment. Then, the leaf vertex is used as a new convex hull node, and the convex hull line is rotated counterclockwise. Repeat this process until the convex hull line touches the existing convex hull connecting line segment. Finally, the figure formed by all convex hull connecting line segments is the convex hull; by accurately connecting the convex hull nodes and leaf vertices, the outline of the plant can be captured more accurately, providing accurate data for plant phenotyping analysis; through continuous rotation and connection operations, all leaves of the plant can be fully covered to ensure the integrity of the convex hull; by constructing the convex hull, it is easier to identify abnormalities in plant growth, which helps to promptly discover and deal with plant health problems; this method dynamically constructs the convex hull based on the actual plant outline, so it can adapt to plants of different shapes and sizes.

[0130] 3. A computer vision-based plant phenotype monitoring method, which segments the convex hull through the growth point and collects data. First, a straight line perpendicular to the horizontal edge is made through the growth point to divide the convex hull into two parts. Then, starting from the first intersection, a straight line is made parallel to the horizontal edge, and a moving threshold is set. The collection line is moved along the first segmentation line to the horizontal edge to obtain the intersection with the convex hull and the segmentation line, and the distance difference between these intersections is calculated, and these differences are entered into the edge difference set. Repeat this process until the growth point is within the collection line. Finally, the minimum and maximum values ​​are selected from the edge difference set to form a reasonable difference interval; by accurately segmenting the convex hull and calculating the distance difference, the growth status and morphological characteristics of the plant can be more accurately evaluated. At the same time, by systematically collecting and analyzing data, the growth of the plant can be fully understood, providing a scientific basis for plant health management, and according to the collection of edge difference sets, rich plant morphological data can be obtained, providing more dimensional information for plant phenotype research.

[0131] 4. A computer vision-based plant phenotypic monitoring method is provided, wherein all potted plants to be detected are entered into a detection set. Then, a potted plant is randomly selected as a detection element, and the growth point and horizontal edge in its side view are obtained. A straight line perpendicular to the horizontal edge is made through the growth point, and the convex hull is divided into a third convex hull and a fourth convex hull. Next, all leaf vertices on the edge contour of the detection convex hull are obtained, and the leaf vertex farthest from the horizontal edge is selected as the canopy point. A straight line parallel to the horizontal edge is made through the growth point, and the vertical distance from the canopy point to the straight line is measured, i.e., the canopy height. If the canopy height of the detection element exceeds the minimum and maximum values ​​in the canopy height set, it is marked as a height anomaly, and the detection element is reselected; if the canopy height is within a reasonable range, the convex hull contour detection strategy is executed; by measuring the canopy height, the growth status of the plant can be accurately monitored, and growth anomalies can be discovered in time. For plants with height anomalies, they can be quickly re-detected to ensure the timeliness and accuracy of the data, and corresponding measures can be taken quickly.

[0132] 5. This is a computer vision-based plant phenotyping monitoring method that identifies abnormal nodes on the convex hull contour of the plant by moving the test line. If an abnormal node is detected, the leaf is judged to be broken by calculating the change in the slope of the tangent line of the leaf contour on the line segment. If the slope change exceeds the preset threshold, the leaf is considered to be broken; otherwise, the plant state is marked as abnormal growth; this leaf break detection method based on slope analysis can effectively distinguish between normal leaf bending and abnormal shape changes caused by breakage. By analyzing the slope change of the tangent line of the leaf contour instead of simple shape or position judgment, the accuracy of detection can be improved and misjudgment can be reduced. This method can perform a more detailed analysis of the abnormal area after detecting the abnormal node, thereby improving the ability to identify broken leaves. In addition, this method systematically processes all leaf vertices on the abnormal line segment to avoid omissions.

[0133] 6. This is a computer vision-based plant phenotyping monitoring method, which forms a closed area by connecting a series of density nodes, and then divides the area into two parts with a dividing line. It calculates the area difference of the corresponding positions in the leaf area sequence sorted from far to near in the two parts. If the area difference between any two corresponding positions exceeds the preset threshold, the plant growth is judged to be abnormal; otherwise, the growth is judged to be normal. This strategy judges whether the plant growth is abnormal by comparing the distribution difference of the leaf area on both sides of the growth point. This method can effectively detect problems such as unilateral growth imbalance during plant growth. By comparing the leaf area in sections instead of simply comparing the overall area, the sensitivity and accuracy of the detection can be improved, and more subtle growth abnormalities can be detected. This method combines the leaf area with the distance from the horizontal edge, takes into account the spatial distribution characteristics of plant growth, and makes the judgment more comprehensive and reliable. In addition, this strategy provides a quantitative evaluation method, avoids the subjectivity of human judgment, and enhances the credibility of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0134] Figure 1 It is a basic flow chart of the present invention;

[0135] Figure 2 is a side view of a potted plant of the present invention;

[0136] Figure 3 This is a schematic diagram of convex hull data collection of the present invention;

[0137] Figure 4 It is a top view of the potted plant of the present invention. DETAILED DESCRIPTION

[0138] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0139] Example 1, see Figures 1 to 4 , a computer vision-based plant phenotype monitoring method, comprising:

[0140] Obtain all potted plants with the same growth cycle and normal growth status, and enter them into a qualified set;

[0141] Potted plants consist of plants and pots, and the states of potted plants include normal growth and abnormal growth;

[0142] The plant phenotypic data collection strategy is sequentially executed on the elements in the qualified set, and the canopy height set and the reasonable difference set are obtained according to the results of the plant phenotypic data collection strategy execution;

[0143] When all elements in the qualified set have completed the plant phenotype data collection strategy, the potted plants that need to be tested are tested in turn;

[0144] If the detection result shows that the canopy height of the detected potted plant is abnormal, the detection is stopped, otherwise, the convex hull detection strategy is executed;

[0145] If an abnormal leaf vertex is found during the execution of the convex hull detection strategy, the slope judgment strategy is executed on it to determine whether the leaf to which the abnormal leaf vertex belongs is broken;

[0146] Determine whether to execute the plant centrifugal strategy according to the execution result of the convex hull contour detection strategy;

[0147] If the plant centrifugal strategy is executed, the potted plant state is set according to the execution result, otherwise, the detection is stopped.

[0148] The plant phenotype data collection strategy is sequentially executed on the elements in the qualified set, and the canopy height set and the reasonable difference set are obtained according to the results of the plant phenotype data collection strategy execution:

[0149] The strategy for collecting plant phenotypic data is as follows:

[0150] Get top and side views of potted plants;

[0151] Among the four edges of the side view, the edge closest to the bottom of the flower pot in the side view is taken as the horizontal edge, and the convex hull nodes are determined according to the horizontal edge and the leaf vertices on the side view;

[0152] Taking the convex hull node determined by the horizontal edge and the leaf node on the side view as the starting point, connect with the remaining leaf nodes in sequence, and record the formed closed graph as the convex hull;

[0153] The point where the bottom of the plant touches the pot in the side view is recorded as the growth point;

[0154] The canopy height is collected for the convex hull according to the growth point, and the collection result is recorded into the canopy height set;

[0155] Draw straight lines through the growth points to segment the convex hull, and collect and process data on the segmented convex hull.

[0156] See also Figure 4 , the edge of the four sides of the side view that is closest to the bottom of the flower pot in the side view is taken as the horizontal edge, and the convex hull nodes are determined according to the horizontal edge and the leaf vertices on the side view:

[0157] Get a top view of the potted plant;

[0158] Mark all leaf vertices in the top view;

[0159] Connect the leaf vertices in the top view in pairs to form multiple connection lines;

[0160] Select the longest connecting line as line segment 1;

[0161] In the top view, draw a straight line perpendicular to line segment 1 through the midpoint of line segment 1, which is recorded as the shooting straight line;

[0162] An image of the potted plant is acquired along a shooting straight line and in a direction pointing toward the potted plant, which is recorded as a side view;

[0163] Randomly select an edge perpendicular to the horizontal edge in the side view as the vertical edge;

[0164] Mark all leaf vertices in the side view;

[0165] Obtain the straight-line distance between each leaf vertex and the horizontal edge in the side view respectively, and enter the leaf vertex with the smallest straight-line distance into the vertex set;

[0166] If the number of leaf vertices in the vertex set is 1, the leaf vertices in the vertex set are used as convex hull nodes;

[0167] If the number of leaf vertices in the vertex set is greater than 1, obtain the straight-line distance between each leaf vertex in the vertex set and the vertical edge, and select the leaf vertex corresponding to the minimum straight-line distance as the convex hull node;

[0168] All leaf vertices are marked through the top view, and the longest connecting line is found as the first line segment. Then, with the midpoint of this line segment as the center, the side view is obtained perpendicular to the line segment direction. By connecting the leaf vertices in the top view two by two, the most suitable angle can be found to comprehensively collect plant phenotypic data, providing complete data support for subsequent analysis. In the side view, a side perpendicular to the horizontal edge is randomly selected as the vertical edge, and all leaf vertices are marked. Next, the straight-line distance from each leaf vertex to the horizontal edge is calculated, and the leaf vertices with the smallest distance are selected to form a vertex set. If there is only one leaf vertex in the set, it is directly used as a convex hull node; if there are multiple, the leaf vertex with the smallest distance to the vertical edge is further selected as the convex hull node; by combining the top view and the side view, the vertex position of the plant leaf can be determined more accurately, thereby improving the accuracy of the convex hull node determination; by selecting the leaf vertex with the smallest distance to the horizontal and vertical edges as the convex hull node, it can better adapt to plants of different shapes and sizes, and improve the robustness of the system.

[0169] The method uses the convex hull node determined by the horizontal edge and the leaf node on the side view as the starting point, connects with the remaining leaf nodes in sequence, and records the formed closed graph as the convex hull, including:

[0170] S1. Draw a straight line passing through the convex hull nodes and parallel to the horizontal edge, recorded as the convex hull straight line;

[0171] S2. If there are leaf vertices other than the convex hull node on the convex hull line, connect the convex hull node and the leaf vertex on the convex hull line that is farthest from the convex hull node to form a convex hull connecting line segment;

[0172] Select the leaf vertex farthest from the convex hull node on the convex hull line as the new convex hull node, and rotate the convex hull line counterclockwise around the convex hull node;

[0173] S3. If the convex hull line only contains convex hull nodes, rotate the convex hull line counterclockwise around the convex hull nodes;

[0174] S4. If the convex hull line passes through the remaining leaf vertices except the convex hull nodes during the rotation process, then perform operations S2 to S4;

[0175] S5. If the convex hull straight line touches the convex hull connecting line segment during the rotation process, the rotation is stopped;

[0176] The figure formed by connecting the line segments of all convex hulls in the side view is recorded as the convex hull;

[0177] By making a convex hull line parallel to the horizontal edge and finding the farthest leaf vertex except the convex hull node on the line, connecting the two points to form a convex hull connecting line segment. Then, take the leaf vertex as the new convex hull node and rotate the convex hull line counterclockwise. Repeat this process until the convex hull line touches the existing convex hull connecting line segment. Finally, the figure formed by all convex hull connecting line segments is the convex hull; by accurately connecting the convex hull nodes and leaf vertices, the outline of the plant can be captured more accurately, providing accurate data for plant phenotyping; through continuous rotation and connection operations, all leaves of the plant can be fully covered to ensure the integrity of the convex hull; by constructing the convex hull, it is easier to identify abnormalities in plant growth, which helps to promptly detect and deal with plant health problems; this method can adapt to plants of different shapes and sizes because it dynamically constructs the convex hull based on the actual plant outline.

[0178] The collecting of the canopy heights of the convex hull according to the growth points and recording the collected results into the canopy height set include:

[0179] Get all leaf vertices on the convex hull edge contour;

[0180] Select the leaf vertex farthest from the horizontal edge as the canopy point;

[0181] Draw a straight line through the growth point parallel to the horizontal edge, record the vertical distance between the canopy point and the straight line as the canopy height, and enter it into the canopy height set.

[0182] See also Figure 3 , the overgrown points are used as straight lines to segment the convex hull, and data collection and processing are performed on the segmented convex hull, including:

[0183] Draw a straight line perpendicular to the horizontal edge through the growth point, which is recorded as the first dividing line;

[0184] The intersection point of the first dividing line and the convex hull is recorded as the first intersection point and the second intersection point;

[0185] The straight-line distance between the first intersection point and the horizontal edge is greater than that between the second intersection point and the horizontal edge;

[0186] Dividing the convex hull into a first convex hull and a second convex hull by a first dividing straight line;

[0187] S6. Draw a straight line through the first intersection point parallel to the horizontal edge, which is recorded as the collection line;

[0188] Set the movement threshold for movement;

[0189] S7, moving the collection line along the first segmentation line toward the horizontal edge, with the moving distance being equal to the moving threshold;

[0190] S8, obtaining the intersection points of the moving collection line with the first convex hull, the second convex hull and the first segmentation line, and recording them as the first convex hull point, the first segmentation point and the second convex hull point respectively;

[0191] The first convex hull point is located at the first convex hull edge contour, and the second convex hull point is located at the second convex hull edge contour;

[0192] S9, calculating the difference between the distance from the first convex hull point to the first segmentation point and the distance from the second convex hull point to the first segmentation point, and recording the difference into an edge difference set;

[0193] Repeating S6 to S9 until the growth point is within the collection line;

[0194] If all elements in the qualified set are traversed, the minimum and maximum values ​​in the edge difference set are selected, and the range between the minimum and maximum values ​​is recorded as the reasonable difference interval;

[0195] Steps to segment the convex hull through the growth point and collect data. First, make a straight line perpendicular to the horizontal edge through the growth point to divide the convex hull into two parts. Then, starting from the first intersection, make a straight line parallel to the horizontal edge and set a moving threshold. Move the collection line along the first segmentation line to the horizontal edge, obtain the intersection with the convex hull and the segmentation line, and calculate the distance difference between these intersections, and enter these differences into the edge difference set. Repeat this process until the growth point is within the collection line. Finally, select the minimum and maximum values ​​from the edge difference set to form a reasonable difference interval; by accurately segmenting the convex hull and calculating the distance difference, the growth status and morphological characteristics of the plant can be more accurately evaluated. At the same time, by systematically collecting and analyzing data, the growth of the plant can be fully understood, providing a scientific basis for plant health management, and according to the collection of edge difference sets, rich plant morphological data can be obtained, providing more dimensional information for plant phenotypic research.

[0196] When all elements in the qualified set have completed the plant phenotype data collection strategy, the potted plants that need to be tested are tested in turn, including:

[0197] Execute a plant phenotypic data collection strategy on the detected potted plants, obtain the canopy height of the detected potted plants according to the collection results, and determine whether the canopy height is within the canopy height set;

[0198] According to the canopy height judgment result, the canopy height status of the detected potted plant is marked to determine whether to execute the convex hull contour detection strategy, as follows:

[0199] Canopy height status includes abnormal height and normal height;

[0200] Get all potted plants that need to be tested and enter them into the test set;

[0201] Randomly select any element in the detection set as the detection element, and obtain the growth point and horizontal edge in the side view of the detection element;

[0202] Draw a straight line perpendicular to the horizontal edge through the growth point, which is recorded as the second dividing line;

[0203] The intersection points of the second segmentation straight line and the edge contour of the detected convex hull are recorded as the third intersection point and the fourth intersection point;

[0204] The straight-line distance between the third intersection point and the horizontal edge is greater than that between the fourth intersection point and the horizontal edge;

[0205] Divide the convex hull into a third convex hull and a fourth convex hull by a second dividing line;

[0206] Get all leaf vertices on the edge contour of the detected convex hull;

[0207] Select the leaf vertex farthest from the horizontal edge as the canopy point;

[0208] Draw a straight line through the growth point parallel to the horizontal edge, and record the vertical distance between the crown point and the straight line as the crown height;

[0209] If the canopy height of the detection element is less than the minimum value in the canopy height set or greater than the maximum value in the canopy height set, the canopy height of the detection element is marked as abnormal height, and any element in the detection set is reselected as a new detection element, and the plant phenotypic data collection strategy is executed;

[0210] If the canopy height of the detected element is greater than or equal to the minimum value in the canopy height set and less than or equal to the maximum value in the canopy height set, the convex hull contour detection strategy is executed;

[0211] All potted plants to be detected are entered into the detection set. Then, a potted plant is randomly selected as the detection element, the growth point and horizontal edge in its side view are obtained, and a straight line perpendicular to the horizontal edge is made through the growth point to divide the convex hull into the third convex hull and the fourth convex hull. Next, all leaf vertices on the edge contour of the detection convex hull are obtained, and the leaf vertex farthest from the horizontal edge is selected as the canopy point. A straight line parallel to the horizontal edge is made through the growth point, and the vertical distance from the canopy point to the straight line is measured, that is, the canopy height. If the canopy height of the detection element exceeds the minimum and maximum values ​​in the canopy height set, it is marked as a height anomaly and the detection element is reselected; if the canopy height is within a reasonable range, the convex hull contour detection strategy is executed; by measuring the canopy height, the growth status of the plant can be accurately monitored, and growth anomalies can be discovered in time. For plants with height anomalies, they can be quickly re-detected to ensure the timeliness and accuracy of the data, and corresponding measures can be taken quickly.

[0212] If the detection result shows that the canopy height of the detected potted plant is abnormal, the detection is stopped; otherwise, the convex hull detection strategy is executed, including:

[0213] The convex hull detection strategy is as follows:

[0214] S10. Draw a straight line through the third intersection point parallel to the horizontal edge, which is recorded as the test line;

[0215] Set the distance threshold used to test the line movement distance;

[0216] S11, moving the test line toward the horizontal edge along the second dividing line, and the moving distance is the distance threshold;

[0217] S12, obtaining the intersection points of the moved test line with the third convex hull, the fourth convex hull and the second segmentation straight line, and recording them as the third convex hull point, the second segmentation point and the fourth convex hull point respectively;

[0218] The third convex hull point is located at the edge contour of the third convex hull, and the fourth convex hull point is located at the edge contour of the fourth convex hull;

[0219] S13, calculating the difference between the distance from the third convex hull point to the second segmentation point and the distance from the fourth convex hull point to the second segmentation point;

[0220] If the above difference is within a reasonable difference range, then execute S14 operation;

[0221] If the difference is not within a reasonable difference range, the third convex hull point and the fourth convex hull point currently corresponding to the difference are marked as abnormal nodes, and operations S11 to S14 are performed;

[0222] S14, in the test line, select the midpoint between the third convex hull point and the second segmentation point, and the midpoint between the fourth convex hull point and the second segmentation point as density nodes;

[0223] The operations S10 to S14 are repeatedly performed until the growth point is on the test line.

[0224] If an abnormal leaf vertex is found during the execution of the convex hull detection strategy, a slope judgment strategy is executed on it to judge whether the leaf to which the abnormal leaf vertex belongs is broken, including:

[0225] Obtain all abnormal nodes on the edge contour of the third convex hull, and select the abnormal nodes closest and farthest from the horizontal edge as interception points respectively;

[0226] The contour edge between the interception points on the third convex hull contour edge is recorded as an abnormal line segment, and the slope judgment strategy is executed as follows:

[0227] Get all leaf vertices on the abnormal line segment and enter them into the abnormal set in descending order according to their distance from the horizontal edge;

[0228] Sequentially obtain the leaf vertices in the exception set as the target leaf vertices;

[0229] Get the leaf contour corresponding to the vertex of the target leaf on the side view, and record it as the target contour;

[0230] The target contour includes an upper contour and a lower contour;

[0231] Taking the target leaf vertex as the starting point, set multiple tangent points with equal intervals on the upper contour in sequence;

[0232] Draw tangent lines tangent to the upper contour through the tangent points respectively, and enter the slope corresponding to each tangent line into the slope set in sequence;

[0233] Set the slope threshold used to determine the slope change;

[0234] The differences of adjacent slopes in the slope set are recorded in turn into the slope change set;

[0235] If there is an element in the slope change set that is greater than the slope threshold, the target node is marked as a normal node, and the leaf vertex on the abnormal line segment is reselected as the new target leaf vertex, and the slope judgment strategy is executed;

[0236] If all elements in the slope change set are less than the slope threshold, the current potted plant state is marked as abnormal growth, and the slope judgment strategy is stopped;

[0237] If all leaf vertices in the abnormal line segment of the third convex hull are traversed and the current potted plant state is not marked as abnormal growth, the abnormal line segment of the fourth convex hull is obtained and the slope judgment strategy is executed;

[0238] By moving the test line, abnormal nodes are identified on the convex hull contour of the plant. If an abnormal node is detected, the leaf is judged to be broken by calculating the change in the slope of the tangent line of the leaf contour on the line segment. If the slope change exceeds the preset threshold, the leaf is considered to be broken; otherwise, the plant state is marked as abnormal growth; this leaf break detection method based on slope analysis can effectively distinguish normal leaf bending from abnormal shape changes caused by breakage. By analyzing the slope change of the tangent line of the leaf contour instead of simple shape or position judgment, the detection accuracy can be improved and misjudgment can be reduced. This method can perform a more detailed analysis of the abnormal area after detecting the abnormal node, thereby improving the recognition ability of leaf breakage. In addition, this method systematically processes all leaf vertices on the abnormal line segment to avoid omissions.

[0239] The step of judging whether to execute the plant centrifugal strategy according to the execution result of the convex hull contour detection strategy includes:

[0240] The plant centrifugation strategy is as follows:

[0241] If the slope judgment strategy shows normal growth, the plant centrifugal strategy is implemented for the potted plants, as follows:

[0242] Randomly select a density node in the side view as the starting point;

[0243] S15, starting from the starting point, in a counterclockwise order, select a density node adjacent to the starting point as the next density node;

[0244] S16, connecting the starting point and the next density node selected;

[0245] S17, taking the next selected density node as a new starting point;

[0246] Repeat steps S16 to S17 until all density nodes are connected to form a closed graph, which is recorded as a density region;

[0247] The second dividing line divides the density area into a first density contrast area and a second density contrast area;

[0248] Set the interval threshold for multiple segmentation line intervals;

[0249] Draw multiple straight lines parallel to the horizontal edge, recorded as density segmentation lines, and the interval between adjacent density segmentation lines is equal to the interval threshold;

[0250] The density segmentation straight line divides the first density contrast region and the second density contrast region into a plurality of sub-density contrast regions respectively;

[0251] Acquire all sub-density comparison areas in the first density comparison area;

[0252] Obtaining the leaf contour area of ​​each sub-density contrast region within the first density contrast region by computer vision;

[0253] According to the order of the distance between the sub-density comparison area and the horizontal edge, the leaf contour area in the first density comparison area is recorded into the first area set from far to near;

[0254] Acquire all sub-density comparison areas in the second density comparison area;

[0255] Obtaining the leaf contour area of ​​each sub-density contrast region within the second density contrast region by computer vision;

[0256] According to the order of the distance between the sub-density comparison area and the horizontal edge, the leaf contour area in the second density comparison area is recorded into the second area set from far to near;

[0257] Set the area threshold used to determine the difference in leaf outline area;

[0258] Select the first element in the first area set and the first element in the second area set respectively, and calculate the absolute value of their difference, which is recorded as the area difference;

[0259] If the area difference is less than or equal to the area threshold, then reselect elements in the first area set and the second area set in turn and calculate the area difference;

[0260] If the area difference is greater than the area threshold, the current potted plant status is marked as abnormal growth;

[0261] If all elements in the first area set and the second area set are traversed, and the area difference corresponding to each element is less than or equal to the area threshold, then the plant growth state of the current detection element is marked as normal growth;

[0262] A closed area is formed by connecting a series of density nodes, and then the area is divided into two parts by a dividing line. It calculates the area difference of the corresponding positions in the leaf area sequence sorted from far to near in the two parts. If the area difference between any two corresponding positions exceeds the preset threshold, the plant growth is judged to be abnormal; otherwise, the growth is judged to be normal. This strategy judges whether the plant growth is abnormal by comparing the distribution difference of the leaf area on both sides of the growth point. This method can effectively detect problems such as unilateral growth imbalance during plant growth. By comparing the leaf area in sections instead of simply comparing the overall area, the sensitivity and accuracy of the detection can be improved, and more subtle growth anomalies can be detected. This method combines the leaf area with the distance from the horizontal edge, takes into account the spatial distribution characteristics of plant growth, and makes the judgment more comprehensive and reliable. In addition, this strategy provides a quantitative evaluation method, avoids the subjectivity of human judgment, and enhances the credibility of the detection results.

[0263] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0264] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A plant phenotype monitoring method based on computer vision, characterized in that: include: Obtain all potted plants with the same growth cycle and normal growth status, and enter them into a qualified set; Potted plants consist of plants and pots, and the states of potted plants include normal growth and abnormal growth; The plant phenotypic data collection strategy is sequentially executed on the elements in the qualified set, and the canopy height set and the reasonable difference set are obtained according to the results of the plant phenotypic data collection strategy execution; When all elements in the qualified set have completed the plant phenotype data collection strategy, the potted plants that need to be tested are tested in turn; If the detection result shows that the canopy height of the detected potted plant is abnormal, the detection is stopped, otherwise, the convex hull detection strategy is executed; If an abnormal leaf vertex is found during the execution of the convex hull detection strategy, the slope judgment strategy is executed on it to determine whether the leaf to which the abnormal leaf vertex belongs is broken; Determine whether to execute the plant centrifugal strategy according to the execution result of the convex hull contour detection strategy; If the plant centrifugal strategy is executed, the potted plant state is set according to the execution result, otherwise, the detection is stopped.

2. A computer vision-based plant phenotype monitoring method according to claim 1, characterized in that: The step of sequentially executing the plant phenotype data collection strategy on the elements in the qualified set, and obtaining the canopy height set and the reasonable difference set according to the results of executing the plant phenotype data collection strategy, includes: The strategy for collecting plant phenotypic data is as follows: Obtain a top view and a side view of the potted plant, take the edge of the four edges of the side view that is closest to the bottom of the flower pot in the side view as a horizontal edge, and determine the convex hull nodes according to the horizontal edge and the leaf vertices on the side view; Taking the convex hull node determined by the horizontal edge and the leaf node on the side view as the starting point, connect with the remaining leaf nodes in sequence, and record the formed closed graph as the convex hull; The point where the bottom of the plant touches the pot in the side view is recorded as the growth point; The canopy height is collected for the convex hull according to the growth point, and the collection result is recorded into the canopy height set; Draw straight lines through the growth points to segment the convex hull, and collect and process data on the segmented convex hull.

3. A computer vision-based plant phenotype monitoring method according to claim 2, characterized in that: The step of obtaining a top view and a side view of the potted plant, taking an edge of the four edges of the side view that is closest to the bottom of the flower pot in the side view as a horizontal edge, and determining a convex hull node according to the horizontal edge and the leaf vertices on the side view, includes: Get a top view of the potted plant; Mark all leaf vertices in the top view; The leaf apex is the tip of each leaf in a potted plant; Connect the leaf vertices in the top view in pairs to form multiple connection lines; Select the longest connecting line as line segment 1; In the top view, draw a straight line perpendicular to line segment 1 through the midpoint of line segment 1, which is recorded as the shooting straight line; An image of the potted plant is acquired along a shooting straight line and in a direction pointing toward the potted plant, which is recorded as a side view; Randomly select an edge perpendicular to the horizontal edge in the side view as the vertical edge; Mark all leaf vertices in the side view; Obtain the straight-line distance between each leaf vertex and the horizontal edge in the side view respectively, and enter the leaf vertex with the smallest straight-line distance into the vertex set; If the number of leaf vertices in the vertex set is 1, the leaf vertices in the vertex set are used as convex hull nodes; If the number of leaf vertices in the vertex set is greater than 1, the straight-line distance between each leaf vertex in the vertex set and the vertical edge is obtained, and the leaf vertex corresponding to the minimum straight-line distance is selected as the convex hull node.

4. The computer vision-based plant phenotype monitoring method according to claim 2, characterized in that: The method uses the convex hull node determined by the horizontal edge and the leaf node on the side view as the starting point, connects with the remaining leaf nodes in sequence, and records the formed closed graph as the convex hull, including: S1. Draw a straight line passing through the convex hull nodes and parallel to the horizontal edge, recorded as the convex hull straight line; S2. If there are leaf vertices other than the convex hull node on the convex hull line, connect the convex hull node and the leaf vertex on the convex hull line that is farthest from the convex hull node to form a convex hull connecting line segment; Select the leaf vertex farthest from the convex hull node on the convex hull line as the new convex hull node, and rotate the convex hull line counterclockwise around the convex hull node; S3. If the convex hull line only contains convex hull nodes, rotate the convex hull line counterclockwise around the convex hull nodes; S4. If the convex hull line passes through the remaining leaf vertices except the convex hull nodes during the rotation process, then perform operations S2 to S4; S5. If the convex hull straight line touches the convex hull connecting line segment during the rotation process, the rotation is stopped; The figure formed by connecting all the convex hull segments in the side view is recorded as the convex hull.

5. The computer vision-based plant phenotype monitoring method according to claim 2, characterized in that: The collecting of the canopy heights of the convex hull according to the growth points and recording the collected results into the canopy height set include: Get all leaf vertices on the convex hull edge contour; The leaf vertex farthest from the horizontal edge is selected as the canopy point; Draw a straight line through the growth point parallel to the horizontal edge, record the vertical distance between the canopy point and the straight line as the canopy height, and enter it into the canopy height set.

6. The computer vision-based plant phenotype monitoring method according to claim 2, characterized in that: The convex hull is segmented by drawing straight lines through the growth points, and data collection and processing are performed on the segmented convex hull, including: Draw a straight line perpendicular to the horizontal edge through the growth point, which is recorded as the first dividing line; The intersection point of the first dividing line and the convex hull is recorded as the first intersection point and the second intersection point; The straight-line distance between the first intersection point and the horizontal edge is greater than that between the second intersection point and the horizontal edge; Dividing the convex hull into a first convex hull and a second convex hull by a first dividing straight line; S6. Draw a straight line through the first intersection point parallel to the horizontal edge, which is recorded as the collection line; Set the movement threshold for movement; S7, moving the collection line along the first segmentation line toward the horizontal edge, with the moving distance being equal to the moving threshold; S8, obtaining the intersection points of the moving collection line with the first convex hull, the second convex hull and the first segmentation line, and recording them as the first convex hull point, the first segmentation point and the second convex hull point respectively; The first convex hull point is located at the first convex hull edge contour, and the second convex hull point is located at the second convex hull edge contour; S9, calculating the difference between the distance from the first convex hull point to the first segmentation point and the distance from the second convex hull point to the first segmentation point, and recording the difference into an edge difference set; Repeating S6 to S9 until the growth point is within the collection line; If all elements in the qualified set are traversed, the minimum and maximum values ​​in the edge difference set are selected, and the range formed between the minimum and maximum values ​​is recorded as the reasonable difference interval.

7. The computer vision-based plant phenotype monitoring method according to claim 1, characterized in that: When all elements in the qualified set have completed the plant phenotype data collection strategy, the potted plants that need to be tested are tested in turn, including: Execute a plant phenotypic data collection strategy on the detected potted plants, obtain the canopy height of the detected potted plants according to the collection results, and determine whether the canopy height is within the canopy height set; According to the canopy height judgment result, the canopy height status of the detected potted plant is marked to determine whether to execute the convex hull contour detection strategy, as follows: Canopy height status includes abnormal height and normal height; Get all potted plants that need to be tested and enter them into the test set; Randomly select any element in the detection set as the detection element, and obtain the growth point and horizontal edge in the side view of the detection element; Draw a straight line perpendicular to the horizontal edge through the growth point, which is recorded as the second dividing line; The intersection points of the second segmentation straight line and the edge contour of the detected convex hull are recorded as the third intersection point and the fourth intersection point; The straight-line distance between the third intersection point and the horizontal edge is greater than that between the fourth intersection point and the horizontal edge; Divide the convex hull into a third convex hull and a fourth convex hull by a second dividing line; Get all leaf vertices on the edge contour of the detected convex hull; The leaf vertex farthest from the horizontal edge is selected as the canopy point; Draw a straight line through the growth point parallel to the horizontal edge, and record the vertical distance between the crown point and the straight line as the crown height; If the canopy height of the detection element is less than the minimum value in the canopy height set or greater than the maximum value in the canopy height set, the canopy height of the detection element is marked as abnormal height, and any element in the detection set is reselected as a new detection element, and the plant phenotypic data collection strategy is executed; If the canopy height of the detected element is greater than or equal to the minimum value in the canopy height set and less than or equal to the maximum value in the canopy height set, the convex hull contour detection strategy is executed.

8. The computer vision-based plant phenotype monitoring method according to claim 1, characterized in that: If the detection result shows that the canopy height of the detected potted plant is abnormal, the detection is stopped; otherwise, the convex hull detection strategy is executed, including: The convex hull detection strategy is as follows: S10. Draw a straight line through the third intersection point parallel to the horizontal edge, which is recorded as the test line; Set the distance threshold used to test the line movement distance; S11, moving the test line toward the horizontal edge along the second dividing line, and the moving distance is the distance threshold; S12, obtaining the intersection points of the moved test line with the third convex hull, the fourth convex hull and the second segmentation straight line, and recording them as the third convex hull point, the second segmentation point and the fourth convex hull point respectively; The third convex hull point is located at the edge contour of the third convex hull, and the fourth convex hull point is located at the edge contour of the fourth convex hull; S13, calculating the difference between the distance from the third convex hull point to the second segmentation point and the distance from the fourth convex hull point to the second segmentation point; If the above difference is within a reasonable difference range, then execute S14 operation; If the difference is not within a reasonable difference range, the third convex hull point and the fourth convex hull point currently corresponding to the difference are marked as abnormal nodes, and operations S11 to S14 are performed; S14, in the test line, select the midpoint between the third convex hull point and the second segmentation point, and the midpoint between the fourth convex hull point and the second segmentation point as density nodes; The operations S10 to S14 are repeatedly performed until the growth point is on the test line.

9. The computer vision-based plant phenotype monitoring method according to claim 1, characterized in that: If an abnormal leaf vertex is found during the execution of the convex hull detection strategy, a slope judgment strategy is executed on it to judge whether the leaf to which the abnormal leaf vertex belongs is broken, including: Obtain all abnormal nodes on the edge contour of the third convex hull, and select the abnormal nodes closest and farthest from the horizontal edge as interception points respectively; The contour edge between the interception points on the third convex hull contour edge is recorded as an abnormal line segment, and the slope judgment strategy is executed as follows: Get all leaf vertices on the abnormal line segment and enter them into the abnormal set in descending order according to their distance from the horizontal edge; Sequentially obtain the leaf vertices in the exception set as the target leaf vertices; Get the leaf contour corresponding to the vertex of the target leaf on the side view, and record it as the target contour; The target contour includes an upper contour and a lower contour; Taking the target leaf vertex as the starting point, set multiple tangent points with equal intervals on the upper contour in sequence; Draw tangent lines tangent to the upper contour through the tangent points respectively, and enter the slope corresponding to each tangent line into the slope set in sequence; Set the slope threshold used to determine the slope change; The differences of adjacent slopes in the slope set are recorded in turn into the slope change set; If there is an element in the slope change set that is greater than the slope threshold, the target node is marked as a normal node, and the leaf vertex on the abnormal line segment is reselected as the new target leaf vertex, and the slope judgment strategy is executed; If all elements in the slope change set are less than the slope threshold, the current potted plant state is marked as abnormal growth, and the slope judgment strategy is stopped; If all leaf vertices in the abnormal line segment of the third convex hull are traversed and the current potted plant state is not marked as abnormal growth, the abnormal line segment of the fourth convex hull is obtained and the slope judgment strategy is executed.

10. The computer vision-based plant phenotype monitoring method according to claim 1, characterized in that: The step of judging whether to execute the plant centrifugal strategy according to the execution result of the convex hull contour detection strategy includes: The plant centrifugation strategy is as follows: If the slope judgment strategy shows normal growth, the plant centrifugal strategy is implemented for the potted plants, as follows: Randomly select a density node in the side view as the starting point; S15, starting from the starting point, in a counterclockwise order, select a density node adjacent to the starting point as the next density node; S16, connecting the starting point and the next density node selected; S17, taking the next selected density node as a new starting point; Repeat steps S16 to S17 until all density nodes are connected to form a closed graph, which is recorded as a density region; The second dividing line divides the density area into a first density contrast area and a second density contrast area; Set the interval threshold for multiple segmentation line intervals; Draw multiple straight lines parallel to the horizontal edge, recorded as density segmentation lines, and the interval between adjacent density segmentation lines is equal to the interval threshold; The density segmentation straight line divides the first density contrast region and the second density contrast region into a plurality of sub-density contrast regions respectively; Acquire all sub-density comparison areas in the first density comparison area; Obtaining the leaf contour area of ​​each sub-density contrast region within the first density contrast region by computer vision; According to the order of the distance between the sub-density comparison area and the horizontal edge, the leaf contour area in the first density comparison area is recorded into the first area set from far to near; Acquire all sub-density comparison areas in the second density comparison area; Obtaining the leaf contour area of ​​each sub-density contrast region within the second density contrast region by computer vision; According to the order of the distance between the sub-density comparison area and the horizontal edge, the leaf contour area in the second density comparison area is recorded into the second area set from far to near; Set the area threshold used to determine the difference in leaf outline area; Select the first element in the first area set and the first element in the second area set respectively, and calculate the absolute value of their difference, which is recorded as the area difference; If the area difference is less than or equal to the area threshold, then reselect elements in the first area set and the second area set in turn and calculate the area difference; If the area difference is greater than the area threshold, the current potted plant status is marked as abnormal growth; If all elements in the first area set and the second area set are traversed and the area difference corresponding to each element is less than or equal to the area threshold, the plant growth state of the current detection element is marked as normal growth.

Images