CN111652885B - Corn seedling stage point cloud stem leaf organ segmentation method - Google Patents
Corn seedling stage point cloud stem leaf organ segmentation method Download PDFInfo
- Publication number
- CN111652885B CN111652885B CN202010348618.2A CN202010348618A CN111652885B CN 111652885 B CN111652885 B CN 111652885B CN 202010348618 A CN202010348618 A CN 202010348618A CN 111652885 B CN111652885 B CN 111652885B
- Authority
- CN
- China
- Prior art keywords
- point
- skeleton
- point cloud
- organ
- points
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2200/00—Indexing scheme for image data processing or generation, in general
- G06T2200/04—Indexing scheme for image data processing or generation, in general involving 3D image data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a corn seedling stage point cloud stem leaf organ segmentation method, which comprises the following steps: 1) Extracting a corn plant point cloud framework; extracting a skeleton of a corn plant point cloud; obtaining a skeleton point set and an undirected graph; 2) Dividing the stem-leaf skeleton, and calculating to obtain a leaf skeleton and a stem skeleton in the organ sub-skeleton; 3) Coarse segmentation of point cloud based on a skeleton; 4) Point cloud fine segmentation based on geodesic distanceIn (a) and (b) point of%p i ) Sequentially dividing into corresponding organ point cloud setsFine segmentation is completed. The technology realizes accurate segmentation of the neo-leaf organs, is particularly effective for the conditions that the neo-leaf blades are close together and the neo-leaf blade base is wrapped, greatly improves the segmentation precision of corn point cloud organs, and provides high-quality organ point cloud data for further corn phenotype detection and organ three-dimensional grid generation.
Description
Technical Field
The invention relates to three-dimensional measurement data processing, in particular to a corn seedling stage point cloud stem leaf organ segmentation method which can be used for corn phenotype detection, corn three-dimensional digital media content (such as three-dimensional animation, games and virtual reality) resource manufacturing and other applications.
Technical Field
In recent years, non-contact measurement of plant plants is becoming a hotspot problem of agriculture and forestry research, wherein three-dimensional point cloud data is increasingly emphasized due to the accuracy of measurement. The three-dimensional point cloud data is used for measuring plant morphological phenotype parameters, the point cloud is required to be segmented in organ scale, and the corn point cloud organ segmentation at present mainly adopts the following methods: (1) skeleton-based segmentation methods. Firstly extracting the skeleton of a corn plant, then dividing the organ skeleton according to skeleton topology, and finally dividing the organ point cloud by utilizing the relation between skeleton points and original point clouds; (2) a machine learning based method. The method learns the characteristics of the organ point cloud through a large number of manual organ point cloud labels, and classifies and segments the point cloud through a constructed model; (3) a region growing-based method. The method is mainly used for segmenting the stem point cloud through a region growing algorithm, and then segmenting each leaf point cloud through region growing from the intersection of the leaf and the stem.
The existing corn point cloud segmentation method only solves the point cloud segmentation problem of corn stem-leaf organs with large leaf spacing, and can not accurately segment seedling stage new leaves. The main reasons are that the pitch between the new leaves is small and the leaves are not fully spread, the leaf bases wrap each other, and thus the dividing error for the new leaves is very large.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention aims to provide the corn seedling stage point cloud stem leaf organ segmentation method, which is used for realizing accurate segmentation of the new-born leaf organ, is particularly effective for the conditions that the new-born leaf is close together and the new-born leaf base is wrapped, greatly improves the segmentation precision of the corn point cloud organ, and provides high-quality organ point cloud data for further corn phenotype detection and organ three-dimensional grid generation.
The invention aims at realizing the following technical scheme, and discloses a method for segmenting a maize seedling stage point cloud stem leaf organ, which comprises the following steps:
1) Corn plant point cloud skeleton extraction
Extracting a skeleton of a corn plant point cloud; obtaining a skeleton point set and an undirected graph;
2) Stem-leaf skeleton segmentation
Obtaining root points, stem root points, leaf tip root points, branch points and joint points in the corn plant skeleton point set;
the root points are skeleton points with only one neighborhood point, the root points at the bottom of the stem are root points at the bottom of the stem, and the root points at the tip of the blade are the rest root points all at the tip of the blade; the branch point is a skeleton point with two neighborhood points, and the join point is a skeleton point with more than two neighborhood points.
Calculating to obtain a leaf sub-skeleton and a stem sub-skeleton in the organ sub-skeleton;
the blade sub-skeleton is the shortest path from each blade tip point to the closest joint point of the topology distance;
the stem skeleton is the shortest path between a root point at the bottom of the stem and a joint point with the farthest topological distance from the root point;
the stem skeleton and all leaf skeletons are organ skeletons [ S ] i (i=1, 2 … … N), where i is the organ and N is the number of organs;
3) Skeleton-based point cloud rough segmentation
Searching K neighbor plant point clouds of each skeleton point in the skeleton point set U in the corn plant point clouds;
a point (point p) in the corn plant point cloud is 3 neighbors of a skeleton point of the i organ, and a category label i is added for the point p;
the point p with only the individual class label i is segmented into the organ i, obtaining a point cloud set of the organ i
The points p with multiple class labels are put into an undivided point cloud setIn (a) and (b);
obtaining a plurality of organ point clouds through point cloud rough segmentationAggregation(i=1, 2 … … n) and an undivided point cloud set +.>
4) Point cloud fine segmentation based on geodesic distance
Will bePoints (p) i ) Sequentially dividing into corresponding organ point cloud sets +.>The specific method for finishing fine segmentation is as follows:
a. computing maize plant point cloud (p) i And p j ) An approximate geodesic distance (d) ij )；
b. For a pair ofThe point clouds of (2) are ordered in the order from the top of the plant to the bottom of the plant, and then the ordered point clouds are ordered from +.>Sequentially taking out points from the set for classification;
c. judging by the following formulaMidpoint p i Which organ (m) belongs to;
said D (p) i M) represents a set point p i To the point ofDistance of>Is the mth organ point cloud set; the p is j Is p i The point is->K (integer between 1 and 20) in the set A, wherein the K neighbor is based on geodesic distance, and the point in the set A is p i K points with nearest geodesic distance between the points;
calculate the point p i Distance to all organ point clouds, wherein the smallest distance is p i Organ to which the subject belongs
d. Point p i From the slaveMiddle move to +.>Updated->Used in the classification operation of the next point; if->If the operation is empty, the whole segmentation operation is ended, otherwise, the step c is continuously executed.
The invention has the beneficial effects that:
(1) The novel blade can be accurately segmented, and the novel blade is particularly effective in the conditions that the novel blades are close together and the novel blade base is wrapped.
(2) The segmentation accuracy of the corn point cloud organ is greatly improved, and high-quality organ point cloud data is provided for further corn phenotype detection and organ three-dimensional grid generation.
(3) By the method, high-quality corn point cloud annotation data can be obtained, and data support is provided for three-dimensional deep learning of corn plants.
Drawings
FIG. 1 is a skeletal diagram of a corn plant point cloud;
FIG. 2 is a rough segmentation result diagram;
fig. 3 shows the point cloud segmentation results at different angles.
Detailed Description
Referring to fig. 1-3, the specific method in this embodiment is as follows: a method for segmenting the stem and leaf organs of a maize seedling stage point cloud comprises the following steps:
1) Corn plant point cloud skeleton extraction
Extracting a skeleton of a corn plant point cloud by a method based on Laplace shrinkage (see reference 1); the generated skeleton mainly comprises the following two types of data information: (1) Set of skeleton points u= { U i -a }; (2) an undirected graph G; see fig. 1, which shows connectivity of skeleton points in the set U, G is represented by a connection matrix of n×n, n being the number of skeleton points;
2) Stem-leaf skeleton segmentation
Obtaining root points, stem root points, leaf tip root points, branch points and joint points in the corn plant skeleton point set;
the root points are skeleton points with only one neighborhood point, the root points at the bottom of the stem are root points positioned at the bottom of the stem, among all the root points, only one root point is a root point (red point) at the bottom of the stem, and the root point at the bottom of the stem is obtained by finding out the root points by adopting a manual interaction method; the blade tip root points (pink points) are the rest root points which are all positioned on the blade tip; the branch point is a skeleton point with two neighborhood points, which is called a branch point (green point), and the joint point is a skeleton point with more than two neighborhood points (blue point).
In the undirected graph G, a Dijkstra method (see reference 2) is adopted to calculate a blade sub-skeleton, wherein the blade sub-skeleton is the shortest path from each blade tip root point to the nearest joint point; each blade sub-skeleton comprises a blade root point, a joint point and a plurality of branch points between the two; the number of the blade root points is consistent with the number of the blade sub-skeletons, and each blade sub-skeleton is skeleton representation of the blade organ point cloud;
after the leaf sub-skeleton is obtained through calculation, a Dijkstra method (see reference 2) is adopted to calculate a stem skeleton, wherein the stem skeleton is the shortest path between a root point at the bottom of a stem and a joint point farthest from the root point in topology; the stem skeleton comprises a stem root point, all joint points and a plurality of branch points;
the stem skeleton and all leaf skeletons are organ skeletons, and the organ skeletons can be S i (i=1, 2 … … N), where i is the organ and N is the number of organs (including 1 stalk organ and N-1 leaf organ), we use S 1 To represent the stem skeleton and the remainder all represent the leaf skeleton; the organ sub-skeleton S i One skeleton point (v) of the organs is a skeleton point of the organs;
document 1.cao, j., tagliasacchi, a., olson, m., zhang, h., and Su, z.,2010.Point cloud skeletons via Laplacian based contraction.In:Proc.IEEE Int.Conf.on Shape Modeling and Applications.Aix-en-Provence,2010, pp.187-197.
Document 2 Dijkstra, E.W.,1959.A note on two problems in connexion with graphs.NumerischeMathematik 1,269-271.
3) Skeleton-based point cloud rough segmentation
After obtaining a leaf skeleton and a stem skeleton, searching 3 neighboring plant point clouds of each skeleton point (v) in the corn plant point clouds in the skeleton point set U, namely for each skeleton point, 3 points closest to the skeleton point in the original plant point cloud;
a point in the corn plant point cloud can be set as a point p, and if the point p is a K neighbor of a skeleton point of the organ i, a category label i is added for the point p;
at least one category label is provided for the points in the plant point cloud, if the points p only have a single category label i, the points p are segmented into organs i and can be usedCome indicatorA point cloud set of officials i;
if point p has multiple class labels, point p is put into an undivided point cloud setIn (a) and (b);
obtaining N organ point cloud sets through point cloud rough segmentation(i=1, 2 … … n) (color point), and an undivided point cloud set +.>(black dot);
4) Point cloud fine segmentation based on geodesic distance
The purpose of the fine segmentation is toThe points in (a) are sequentially segmented into corresponding organ point cloud sets +.>The whole process is as follows:
step 1: the geodesic distance between the corn plant point clouds can be calculated by a method based on the Sinkhorn distance (document 3), and any two points in the corn plant point cloud can be respectively set as p i And p j The p is i And p j Symbol d for geodesic distance between ij A representation;
step 2, pairThe points of (a) are ordered in the order from the top of the plant to the bottom of the plant, i.e. the higher the ground the earlier the point ordering, the closer to the ground the later the point ordering, and then from the ordered +.>Sequentially taking out points from the collection for dividingClass;
step 3, fromTake out a point p i Judging which organ the point belongs to, and setting the point cloud set of the mth organ as +.>[ i.e. ] A.C>(i=m), i.e. setpoint p i To->Distance D (p) i M) is obtained according to the following formula:
wherein: p is p j Is p i At the point ofA point in a K (integer between 1 and 20) neighbor set A, wherein the K (integer between 1 and 20) neighbor is based on geodesic distance, and the point in the set A is p i K (integer between 1 and 20) points with nearest geodesic distance between points, the parameter K (integer between 1 and 20) being specified by the user, if the organ set +.>The number of the point clouds in the system is smaller than K (an integer between 1 and 20), and the value of K (an integer between 1 and 20) is +.>The number of the point clouds in the network; calculate the point p i Distance to all organ point clouds, wherein the smallest distance is p i The organ is treated with->A representation;
step 4, point p i From the slaveMiddle move to +.>Updated->Used in the classification operation of the next point (return to step 3); if->If the operation is empty, ending the whole segmentation operation, otherwise, continuing to execute the step 3);
fig. 3 shows the point cloud segmentation result after fine segmentation, and fig. 3 shows the point cloud segmentation from different angles, and it can be seen from the figure that the nascent She Dian cloud is closely adjacent to each other and the leaflet is wrapped by the leaflet, so that the method can accurately segment the leaflet.
Document 3: cuturi M.sink horn Distances Lightspeed Computation of Optimal Transportation Distances [ J ]. Advances in neural information processing systems,2013,26:2292-2300.
Claims (1)
1. A method for segmenting the stem and leaf organs of a maize seedling stage point cloud comprises the following steps:
1): corn plant point cloud skeleton extraction
Extracting a skeleton of a corn plant point cloud; obtaining a corn plant skeleton point set and an undirected graph;
2): stem-leaf skeleton segmentation
Obtaining root points, stem root points, leaf tip root points, branch points and joint points in the corn plant skeleton point set;
the root points are skeleton points with only one neighborhood point, the root points at the bottom of the stem are root points at the bottom of the stem, and the root points at the tip of the blade are the rest root points all at the tip of the blade; the branch point is a skeleton point with two neighborhood points, and the joint point is a skeleton point with more than two neighborhood points;
calculating to obtain a leaf sub-skeleton and a stem sub-skeleton in the organ sub-skeleton;
the blade sub-skeleton is the shortest path from each blade tip point to the closest joint point of the topology distance;
the stem skeleton is the shortest path between a root point at the bottom of the stem and a joint point with the farthest topological distance from the root point;
the stem skeleton and all leaf skeletons are organ skeletons S i (i=1, 2 … … N), i being the organ and N being the number of organs;
3): skeleton-based point cloud rough segmentation
Searching 3 neighboring plant point clouds of each skeleton point in the skeleton point set U in the corn plant point clouds;
a point p in the corn plant point cloud is 3 adjacent to a skeleton point of the organ i, and a category label i is added for the point p;
the point p with only the individual class label i is segmented into the organ i, obtaining a point cloud set of the organ i
The points p with multiple class labels are put into an undivided point cloud setIn (a) and (b);
obtaining a plurality of organ point cloud sets through point cloud rough segmentationAn undivided point cloud set4): point cloud fine segmentation based on geodesic distance
Will bePoint p in (a) i Sequentially dividing into corresponding organ point cloud sets +.>The specific method for finishing fine segmentation is as follows:
a. calculating any two points p in corn plant point cloud i And p j Approximate geodesic distance d between ij ；
b. For a pair ofThe point clouds of (2) are ordered in the order from the top of the plant to the bottom of the plant, and then the ordered point clouds are ordered from +.>Sequentially taking out points from the set for classification;
c. judging by the following formulaMidpoint p i Which organ m belongs to;
said D (p) i M) represents a set point p i To the point ofDistance of>Is the mth organ point cloud set; the point p j Is p i At the position ofK nearest neighbor in (a)A point in the set A, K is an integer between 1 and 20, the K neighbor is based on geodesic distance, and the point in the set A is a point p i K points with the nearest distance between the ground wires;
calculate the point p i Distance to all organ point clouds, wherein the smallest distance is p i Point cloud collection of organs
d. Point p i From the slaveMiddle move to +.>Updated->Used in the classification operation of the next point; if->If the operation is empty, the whole segmentation operation is ended, otherwise, the step c is continuously executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010348618.2A CN111652885B (en) | 2020-04-28 | 2020-04-28 | Corn seedling stage point cloud stem leaf organ segmentation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010348618.2A CN111652885B (en) | 2020-04-28 | 2020-04-28 | Corn seedling stage point cloud stem leaf organ segmentation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111652885A CN111652885A (en) | 2020-09-11 |
| CN111652885B true CN111652885B (en) | 2023-08-18 |
Family
ID=72348458
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010348618.2A Active CN111652885B (en) | 2020-04-28 | 2020-04-28 | Corn seedling stage point cloud stem leaf organ segmentation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111652885B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114581626A (en) * | 2022-01-17 | 2022-06-03 | 北京市农林科学院信息技术研究中心 | Plant three-dimensional phenotype analysis method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103065352A (en) * | 2012-12-20 | 2013-04-24 | 北京农业信息技术研究中心 | Plant three-dimensional reconstruction method based on image and scanning data |
| CN110853044A (en) * | 2019-04-25 | 2020-02-28 | 华中农业大学 | Potted corn point cloud rapid segmentation method based on conditional Euclidean clustering |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11188752B2 (en) * | 2018-03-08 | 2021-11-30 | Regents Of The University Of Minnesota | Crop biometrics detection |
-
2020
- 2020-04-28 CN CN202010348618.2A patent/CN111652885B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103065352A (en) * | 2012-12-20 | 2013-04-24 | 北京农业信息技术研究中心 | Plant three-dimensional reconstruction method based on image and scanning data |
| CN110853044A (en) * | 2019-04-25 | 2020-02-28 | 华中农业大学 | Potted corn point cloud rapid segmentation method based on conditional Euclidean clustering |
Non-Patent Citations (1)
| Title |
|---|
| 玉米果穗点云分割方法研究;温维亮;郭新宇;杨涛;赵德达;苗腾;朱宏宇;董成玉;;系统仿真学报(第12期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111652885A (en) | 2020-09-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | The persistent homology mathematical framework provides enhanced genotype-to-phenotype associations for plant morphology | |
| CN112257597B (en) | Semantic segmentation method for point cloud data | |
| Miao et al. | Automatic stem-leaf segmentation of maize shoots using three-dimensional point cloud | |
| CN110992473B (en) | Tree branch modeling method and system based on vehicle-mounted laser scanning point cloud | |
| CN106780458B (en) | Point cloud framework extraction method and device | |
| CN102903145B (en) | Method for three-dimensional reconstruction of plant population morphological structure | |
| CN110378923A (en) | A kind of method and apparatus that the segmentation of intelligence intratracheal tree is extracted and is classified | |
| CN109064471A (en) | A kind of three-dimensional point cloud model dividing method based on skeleton | |
| Sun et al. | High resolution 3D terrestrial LiDAR for cotton plant main stalk and node detection | |
| Miao et al. | Label3DMaize: toolkit for 3D point cloud data annotation of maize shoots | |
| CN111652885B (en) | Corn seedling stage point cloud stem leaf organ segmentation method | |
| Chaudhury et al. | 3D plant phenotyping: All you need is labelled point cloud data | |
| Zhu et al. | A calculation method of phenotypic traits based on three-dimensional reconstruction of tomato canopy | |
| Zhu et al. | A method for detecting tomato canopies’ phenotypic traits based on improved skeleton extraction algorithm | |
| Miao et al. | A single plant segmentation method of maize point cloud based on Euclidean clustering and K-means clustering | |
| Xu et al. | An improved space colonization algorithm with DBSCAN clustering for a single tree skeleton extraction | |
| CN112991300B (en) | Single wood skeleton extraction and visualization method based on neighborhood characteristics | |
| CN103337092A (en) | An extraction method for a fruit tree limb skeleton | |
| Wang et al. | DFSP: A fast and automatic distance field-based stem-leaf segmentation pipeline for point cloud of maize shoot | |
| Zuo et al. | A method of crop seedling plant segmentation on edge information fusion model | |
| Ma et al. | A method for calculating and simulating phenotype of soybean based on 3D reconstruction | |
| CN113111830A (en) | Grape vine winter pruning point detection algorithm | |
| CN108229502A (en) | A kind of three dimensional point cloud middle period chip base point extracting method of crown canopy | |
| Sodhi et al. | Robust plant phenotyping via model-based optimization | |
| Wen et al. | Accurate and semantic 3D reconstruction of maize leaves |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |