CN103489192B - Method for detecting number of Arabidopsis leaves and distance between cusp and center of mass of each leaf - Google Patents

Abstract

A method for detecting the number of Arabidopsis leaves and the distance between the cusp and the center of mass of each leaf particularly comprises the steps that a demarcating board is placed in a planting pot of Arabidopsis and RGB images of the Arabidopsis are collected through a camera; the collected images are pre-processed, so that the images are automatically corrected and demarcated, wherein the images are corrected so that distortion of the images can be corrected and the images are demarcated so that the real dimension of a unit pixel can be obtained; the pre-processed images are divided, the Arabidopsis is separated from backgrounds and extracted from the images; after Arabidopsis images are obtained through division, the number of the Arabidopsis leaves is extracted, and the distance between the cusp and the center of mass of each leaf is calculated. A phenotype parameter referring to the distance between the cusp and the center of mass of each leaf is used for reflecting the size of each leaf, and the growth situation can be described in a quantitative mode according to the number of the leaves of the Arabidopsis and the size of each leaf. Differences of the growth situations of the leaves of the Arabidopsis with different genes are described through the phenotype parameters so that the functions of the different genes and influence on the Arabidopsis by the genes can be inferred.

Description

A detection method for the number of Arabidopsis thaliana leaves and the distance from the tip of the leaf to the center of mass

technical field

The invention relates to a method for detecting the number of leaves of Arabidopsis thaliana and the distance from the apex of the leaves to the center of mass. A camera is used to collect pictures of Arabidopsis thaliana, and the number of leaves of Arabidopsis thaliana and the tip of the leaves are extracted from the image by using an image processing method. The distance from the point to the centroid enables non-destructive testing of the growth process of Arabidopsis thaliana.

Background technique

Arabidopsis is an important model plant in botany, genetics and genetics. The research on Arabidopsis phenotype can comprehensively and thoroughly clarify the physiological function of Arabidopsis thaliana, especially the relationship between its phenotype and its genes, and the influence of different environmental conditions on its growth. The detection methods of plant phenotypic characteristics include destructive measurement, contact measurement and computer vision detection methods. Destructive measurement is to randomly select a certain number of plants for a batch of plants, and measure their parameters with a destructive method. Contact measurements are measurements of plant parameters using contact sensors. Using computer vision technology to measure is to obtain the spectral image of the measured object through related equipment, including CCD camera, light source, etc., and use related software and algorithms to process the image to obtain the required data, so as to obtain the phenotypic parameters of the plant . Existing research work mainly uses computer vision technology to realize the analysis of individual leaves or other crops. Li Xinguo et al. used a scanner to obtain the images of rapeseed leaves, and used Photoshop software to obtain the number of leaf pixels, and obtained the area of the leaves through the resolution (Li Xinguo, Cai Shengzhong, Li Shaopeng, etc. Determination of avocado leaf area by digital image technology [J]. Tropical Agriculture Science, 2009, 29(2):10-13.). Han Dianyuan et al. proposed an algorithm for segmenting leaves under a white background by color, and then used the reference rectangular plate in the background to calculate the area of the leaves (Han Dianyuan, Huang Xinyuan, Fu Hui, etc. Plant leaf segmentation based on color channel similarity image segmentation method Area calculation [J]. Journal of Agricultural Engineering, 2012, 28(6): 179-183.). Li Shaokun and others use image technology to collect images of corn and wheat, and extract relevant parameters (Li Shaokun, Zhang Xian. Research on multimedia image processing technology for crop plant type information [J]. Acta Crops, 1998, 24(3): 265-271 ). Li Changying et al. used computer vision technology to monitor the growth of greenhouse plants non-destructively, and obtained the external morphological characteristics of plants, including the projected area of leaf crowns and plant height (Li Changying, Teng Guanghui, Zhao Chunjiang, etc.). Using computer vision technology to realize the growth of greenhouse plants Non-destructive monitoring of [J]. Agricultural Engineering Journal, 2003, 19(3): 140-143.).

In summary, the existing research has the following shortcomings:

1. The use of destructive measurement methods will cause damage to plants, and continuous measurement of plant growth cannot be performed.

2. The use of sensors for measurement and direct contact with plants will have a certain impact on the growth of plants, and its cost is high, and its development is relatively difficult.

3. The existing computer vision technology mainly focuses on the phenotype detection of a single leaf or other crops, while the phenotype detection of Arabidopsis mainly relies on manual implementation, which has a large workload and low efficiency.

At present, there are few literature reports on computer vision detection of Arabidopsis phenotypes at home and abroad.

Contents of the invention

The technical problem to be solved by the present invention is: how to use computer vision technology to perform non-destructive testing on Arabidopsis thaliana according to the characteristics of Arabidopsis thaliana, and extract the parameters of the number of leaves in the growth process and the distance from the tip of the leaf to the center of mass. The distance from the tip of each blade to the centroid is used to reflect the size of each blade. The number of leaves of a plant and the size of each leaf can not only quantitatively describe the growth of Arabidopsis thaliana, but also be used in the study of Arabidopsis gene function, that is, through these phenotypic parameters to describe the different genes in Arabidopsis thaliana. The differences in growth conditions can infer the functions of different genes and their influence on Arabidopsis plants.

(1) Technical solution

In order to achieve the above object, the invention provides the number of Arabidopsis thaliana leaves based on computer vision and the detection method of the distance from the tip of the leaf to the center of mass, comprising the following steps:

S1. In the planting pot of Arabidopsis thaliana, place a calibration plate, and use the camera to collect RGB images of Arabidopsis thaliana;

S2. Preprocessing the collected image to realize automatic correction and calibration of the image, wherein the image correction is to correct the distortion of the image, and the image calibration is to obtain the true size of the unit pixel;

S3. Segment the preprocessed image, segment the Arabidopsis thaliana from the background, and extract it from the image;

S4. Extract the number of leaves of Arabidopsis thaliana, and calculate the distance from the tip of each leaf to the centroid.

The calibration plate in the collected image is a black and white checkerboard with a blue frame, which consists of 3×3 squares with a side length of 4mm. Place it on the side of the Arabidopsis plant and perform image acquisition with it.

Step S2 specifically includes the following steps:

S2.1 Position black and white checkerboard

S2.1.1 Extract the blue border according to the blue RGB feature;

S2.1.2 fill the hole inside the image, and then subtract the original blue border image to obtain a new image;

S2.1.3 Perform an open operation on the new image to remove noise points; then perform a close operation to connect the breakpoints to obtain a black and white checkerboard area;

S2.2 Corner detection

S.2.2.1 Calculate the horizontal and vertical first derivatives of each point in the black and white checkerboard area to obtain three new images: the square of the horizontal first derivative, the square of the vertical first derivative and two ones the product of derivatives;

S.2.2.2 Use Gaussian filtering to filter the three images to remove noise;

S.2.2.3 Compose the correlation matrix from the above three images, calculate the criterion function, and judge whether the pixel is a corner point;

S2.3 Image correction and image calibration

S.2.3.1 Obtain the coordinates of the vertices of each square of the checkerboard in the image through corner detection, and obtain the transformation matrix of the two according to their spatial position relationship in the real world;

S.2.3.2 Calculate the inverse of the transformation matrix and act on the image to realize image correction;

S2.3.3 Obtain the total number of pixels of the black and white checkerboard through the coordinates of the corner points, and obtain the actual size of the unit pixel according to its actual size;

Step S3 specifically includes the following steps:

S3.1 Initial image segmentation

S3.1.1 normalize the RGB value of each pixel to obtain rgb, extract the color difference map of 3g-2.4r-b, and use 0 as the threshold to binarize the image;

S3.1.2 Subtract the original detected blue border area from the result, and judge whether the gray value of G of the obtained foreground area pixel is greater than 50, if so, keep it, otherwise remove it;

S3.1.3 Extract the area with the most pixels in the foreground connected area, which is the plant area;

S3.2 Remove noise

S3.2.1 Perform an open operation on the obtained image to obtain a new image, which will only retain the large leaf area and the center area of the plant, and remove the details in the image, including the noise around the stem of the plant and the leaves;

S3.2.2 Calculate the number of pixels of each connected region of the removed detail part, and directly remove the number of pixels less than 12;

S3.2.3 The connected regions with the number of pixels greater than or equal to 12 are separately superimposed with the new image obtained by the open operation, and then calculate the number of regions in the superimposed image. If the number of regions decreases, it means that the connected region is a stem and must be retained , otherwise, if the number of regions in the image increases or remains unchanged after superposition, it means that the connected region is the noise around the leaf and must be removed;

S3.2.4 After judging all the details, superimpose all the retained connected regions with the new image obtained by the open operation to obtain the final plant region;

Step S4 specifically includes the following steps:

S4.1 fills the holes in the plant area, and calculates the centroid of the area;

S4.2 Extract the outer contour points of the filled plant area through the contour search algorithm;

S4.3 calculates the distance from each outer contour point to the centroid of the plant area, thereby turning the 2-dimensional (2D) image into a 1-dimensional (1D) signal;

S4.4 Perform data point image filling on the 1D signal described in S4.3 to expand it to an integer power of 2 to obtain a new 1D signal;

S4.5 performs 4-layer Haar wavelet decomposition on the 1D signal obtained in S4.4, and extracts the wavelet coefficients of the fourth layer having the same length as the 1D signal described in S4.3;

S4.6 Find the positive zero-crossing points in the wavelet coefficients, that is, the previous point of a certain point is less than 0 and the next point is greater than 0. These positive zero-crossing points represent the local maximum of the 1D signal described in S4.3; for For a single leaf, the tip of the leaf is the farthest from the centroid of the plant area, that is, the local maximum of the 1D signal is the cusp of each leaf, so the number of leaves of the plant can be obtained by finding the local maximum of the 1D signal;

S4.7 Calculate the length from the cusp to the centroid of each leaf in the image, and multiply the number of pixels of the length by the real length of the unit pixel obtained in the image calibration step to obtain the real distance from the cusp to the centroid of each leaf.

(2) Beneficial results

The method of the invention uses computer vision technology to collect images of Arabidopsis thaliana, and uses image processing technology to realize the non-destructive detection of Arabidopsis phenotype. By detecting the blue border of the black and white checkerboard of the blue frame, and using the corner detection algorithm to detect the internal corner points, the automatic correction and calibration of the image is finally realized. For Arabidopsis thaliana, the linear combination of normalized rgb values is used for threshold segmentation, and by judging the details of the plant area again, unnecessary noise is removed, so that the Arabidopsis can be divided without affecting the normal growth of Arabidopsis. Mustard is separated from the complex natural growth environment. After the plants were segmented, the number of Arabidopsis leaves and the distance from the tip of each leaf to the centroid were extracted. Compared with traditional manual observation and measurement, this method has greatly improved the efficiency.

Description of drawings

Other characteristics, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

Fig. 1 is the flow chart of the distance detection method of the number of Arabidopsis thaliana leaves and the blade apex to the centroid according to one embodiment of the present invention;

Fig. 2 is used in the method provided by the present invention to be used for the acquisition transposition of acquisition image;

Fig. 3 is the image generated in the process of the preprocessing method provided by the present invention, wherein, the subgraph (1) is the RGB image of Arabidopsis thaliana collected according to the method of the present invention, and (2) is the black and white checkerboard generated after preprocessing grid area, (3) is the corner detection image generated after preprocessing, and (4) is the corrected image generated after preprocessing;

Fig. 4 is the image generated during the processing of the image segmentation method provided by the present invention, wherein, the sub-image (1) is the binary image generated after the initial image segmentation step, and (2) is the final image generated after the noise removal step image of Arabidopsis;

Fig. 5 is the number of leaves of Arabidopsis thaliana provided by the present invention and the image generated in the process of extracting the distance from the tip of the leaf to the center of mass, wherein, the sub-figure (1) is the result after the holes in the plant area are filled, ( 2) is the outer contour of the plant area extracted by the contour search algorithm, (3) is the distance signal from each outer contour point to the centroid, (4) is the result obtained by expanding the length of the distance signal to an integer power of 2, (5) is the fourth layer wavelet coefficient obtained after the distance signal is decomposed by Haar wavelet, and its positive zero-crossing point, (6) is the detected local maximum value of the original distance signal, (7) is the detected blade tip points, and the distance from each cusp to the centroid.

detailed description

A method for detecting the number of Arabidopsis thaliana leaves and the distance from the apex of the leaves to the center of mass proposed by the present invention is described in detail in conjunction with the accompanying drawings and embodiments as follows. The following examples will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be pointed out that those skilled in the art can make several modifications and changes without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

In order to quickly extract the phenotypic parameters of Arabidopsis thaliana, the invention discloses a detection method for the number of Arabidopsis thaliana leaves and the distance from the tip of the leaves to the center of mass. After the method preprocesses the collected images, Arabidopsis is segmented from the complex background environment. On this basis, the parameters of the number of leaves of Arabidopsis and the distance from the tip of the leaf to the center of mass are extracted, which improves the accuracy of the simulation. Acquisition efficiency of phenotypic parameters in A. thaliana.

As shown in Figure 1, according to an embodiment of the present invention, the number of Arabidopsis thaliana leaves and the distance detection method from the tip of the leaf to the centroid include steps:

S1 In the planting pot of Arabidopsis thaliana, place a calibration plate, and use a CCD camera to collect RGB images of Arabidopsis thaliana;

In this example, the acquisition device in FIG. 2 may be used for image acquisition. The collection device includes: a CCD camera 1, a bracket 2, an illumination device 3 and Arabidopsis 4, and a calibration plate 5 is placed on one side of the Arabidopsis. The calibration board uses a black and white checkerboard with a blue frame, which consists of 3×3 squares with a side length of 4mm.

S2 preprocesses the collected images, specifically including the following sub-steps:

S2.1 Position black and white checkerboard

S2.1.1 According to the RGB characteristics of blue, the gray values of R and G are both smaller than B, and the gray value of B is greater than 150, and the blue border is segmented from the image.

S2.1.2 Fill the hole inside the image, and subtract the original blue border image from the result to obtain a new image.

S2.1.3 Perform an opening operation on the new image to remove noise points; then perform a closing operation to connect the breakpoints to obtain a black and white checkerboard area.

S2.2 Corner detection

S.2.2.1 Calculate the horizontal and vertical first derivatives of each point in the black and white checkerboard area. The following Prewitt template is used for calculation, that is, the difference between the third column and the first column of the 3×3 area is used to approximate the derivative in the horizontal direction, and the difference between the third row and the first row is used to approximate the derivative in the vertical direction. Use these two sub-templates to convolve with the image to get two matrices with the same size as the image, denoted as I _x , I _y , and then calculate three new images: I _x ² , I _y ² and I _x I _y .

S.2.2.2 Considering that the image will be disturbed by noise, use a 101×101 Gaussian window to filter the three images to remove noise.

S.2.2.3 The correlation matrix M is composed of three new images:

$\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\left[\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}& {\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}\\ {\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}& {\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\end{array}\right]$

Use this matrix to calculate the criterion function R:

R=det(M)－k·(trace(M)) ²

Among them, k is generally taken as 0.04.

For each pixel point, an R value can be obtained. If the R value of a point is greater than 0.01R _max and it is the local maximum value of the 3×3 neighborhood, it will be judged as a corner point.

S2.3 Image correction and image calibration

S.2.3.1 Obtain the coordinates of the vertices of each square of the checkerboard in the image through corner point detection. In the real world, these vertices form each square, and the coordinates of these vertices can be set according to their spatial position relationship. Let the coordinates of a point in the image be (x', y') ^T , and the coordinates in space be (x, y) ^T , and its homogeneous coordinates are (x' ₁ , x' ₂ , x' ₃ ) ^T and (x ₁ , y, 1) ^T , there is a projective distortion transformation between them, and the transformation matrix is a linear transformation H about the homogeneous three-dimensional coordinates, expressed as follows:

$\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{ccc}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}& {\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 12</span>}}& {\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 13</span>}}\\ {\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}& {\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; two</span>}}& {\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; three</span>}}\\ {\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 31</span>}}& {\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 32</span>}}& {\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 33</span>}}\end{array}\right]\left[\begin{array}{c}\mathrm{<span\; class="notranslate">\; x</span>}\\ \mathrm{<span\; class="notranslate">\; the\; y</span>}\\ \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right]$

Because the non-homogeneous coordinates of the image can be expressed by the homogeneous coordinates as:

${\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 12</span>}}\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 13</span>}}}{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 31</span>}}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 32</span>}}\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 33</span>}}}$

${\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; two</span>}}\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; three</span>}}}{{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 31</span>}}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 32</span>}}\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 33</span>}}}$

So each set of matching points can get two following equations:

(h ₃₁ x+h ₃₂ y+h ₃₃ )x′=h ₁₁ x+h ₁₂ y+h ₁₃

(h ₃₁ x+h ₃₂ y+h ₃₃ )y′=h ₂₁ x+h ₂₂ y+h ₂₃

The transformation matrix H can be obtained with at least four pairs of matching points.

S.2.3.2 Obtain the inverse of the transformation matrix H, and then act on the image to realize image correction.

S2.3.3 Calculate the total number of pixels N _b in the black and white checkerboard area through the coordinates of the corner points, and according to its real size, that is, 9×(4mm) ² =144mm ² , to obtain the real size of the unit pixel, namely:

${\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; dz</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 144</span>}}{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}}$

S3 segments the preprocessed image, and extracts Arabidopsis from the image, specifically including the following sub-steps:

S3.1 Initial image segmentation

S3.1.1 Normalize the RGB value of each pixel to obtain rgb, as follows:

$\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; R</span>}}{\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; B</span>}}\\ \mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; G</span>}}{\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; B</span>}}\\ \mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; B</span>}}{\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; B</span>}}\end{array}\right.$

Extract the color difference map of 3g-2.4r-b, and use 0 as the threshold to binarize the image, that is: if

I=3g－2.4r－b≥0

Then the pixel is judged as the pixel of the foreground plant, otherwise it is the background pixel.

S3.1.2 Subtract the original detected blue frame area from the result. Determine whether the gray value of G of the pixel in the foreground area is greater than 50, if so, keep the pixel, otherwise determine the pixel as a background pixel and remove it from the area.

S3.1.3 Extract the area with the most pixels in the foreground connected area, which is the plant area.

S3.2 Remove noise

S3.2.1 Use the 7×7 disk template to open the obtained image to obtain a new image, which will only retain the large leaf area and the center area of the plant, and remove the detailed parts of the image, including the stems and leaves of the plant of noise.

S3.2.2 Calculate the number of pixels in each connected region of the removed detail part, and remove the number of pixels less than 12 directly.

S3.2.3 The connected regions with the number of pixels greater than or equal to 12 are separately superimposed with the new image obtained by the open operation, and then calculate the number of regions in the superimposed image. If the number of regions decreases, it means that the connected region is a stem and must be retained , otherwise, if the number of regions in the image increases or remains unchanged after superimposition, it means that the connected region is the noise around the leaf and must be removed.

S3.2.4 After judging all the details, superimpose all the retained connected regions with the new image obtained by the open operation to obtain the final plant region.

S4 extracts the number of leaves of Arabidopsis thaliana, and calculates the distance from the tip of each leaf to the centroid, specifically including the following sub-steps:

S4.1 Hole filling

S4.1.1 A point in the hole is given as the starting point, and it is set to 1, and all other points are set to 0, forming a new area X ₀ .

S4.1.2 Use the following symmetrical structure unit to expand X ₀ , and perform intersection operation with the complement of the original image area to obtain a new area X ₁ .

S4.1.3 Determine whether X ₁ is equal to X ₀ , if they are equal, the hole filling ends, otherwise set X ₀ =X ₁ , and repeat the second step.

S4.2 Find contours

S4.2.1 Select the point at the upper left corner of the plant area after filling the hole as the starting point b ₀ , let c ₀ be the background point on the west side of b ₀ .

S4.2.2 Starting from c ₀ , detect 8 adjacent points of b ₀ in a clockwise direction, let b ₁ be the first detected point that is not 0, and let c ₀ be the point before b ₁ in the detection sequence , store b ₀ , b ₁ .

S4.2.3 Determine whether b ₁ is equal to b ₀ , if they are equal, the contour search ends, otherwise set b ₀ =b ₁ , and repeat the second step.

S4.3 Extract 1D signal

S4.3.1 After searching the contour, obtain the contour sequence, and extract the x-coordinate and y-coordinate of each contour point in the contour sequence.

S4.3.2 Calculate the distance from each outer contour point to the centroid of the plant area. Let the coordinates of a certain outer contour point in the image be (x _i , y _i ), the coordinates of the center of mass in the image be (x _o , y _o ), and the distance between them is:

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

By calculating the distance D _i of each point, a 1D signal S is obtained, thereby turning the 2D image into a 1D signal.

S4.4 Extended signal

S4.4.1 Let the length of the 1D signal described in S4.3 be N ₀ , calculate the base 2 logarithm of N ₀ , round the obtained result to an integer and add 1 to obtain the value k.

S4.4.2 If 2 ^k is less than 2N ₀ , perform mirror flip on the 1D signal described in S4.3, supplement the 1D signal described in S4.3, and take the first 2 ^k data points to obtain a new signal .

S4.4.3 If 2 ^k is greater than 2N ₀ , after supplementing the 1D signal described in S4.3 with the mirror-inverted signal, supplement the 1D signal described in S4.3, and then take the first 2 ^k data points, get a new signal.

S4.5 Wavelet transform

S4.5.1 Perform 4-layer Haar wavelet decomposition on the 1D signal obtained in S4.4, and extract the wavelet coefficients of the 4th layer.

S4.5.2 Extract the first N ₀ data points from the obtained wavelet coefficients to obtain a new 1D signal S _C , so that the 1D signals described in S4.3 have the same length.

S4.6 Find the positive zero-crossing point

S4.6.1 Shift the 1D signal S _C to the right and to the left by one bit respectively to obtain S _C1 and S _C2 .

S4.6.2 Find points satisfying the following conditions:

(S _C =0I S _C2 >0)∪(S _C1 <0I S _C <0I S _C2 >0)

Points that meet the above conditions are the positive zero-crossing points of the S _C signal, that is, the previous point of a certain point is less than 0 and the next point is greater than 0. These positive zero-crossing points represent the local part of the 1D signal S described in S4.3. maximum value. For a single leaf, its leaf cusp is farthest from the centroid of the plant area, ie the local maximum of the 1D signal is the cusp of each leaf. Therefore, the number of leaves of the plant can be obtained by finding the local maximum value of the 1D signal.

S4.7 Blade length

S4.7.1 Calculate the length from the cusp of each blade in the image to the centroid, let the coordinates of a leaf cusp in the image be ( _xi , y _i ), and the coordinates of the centroid in the image be (x _o , y _o ), the distance between them is:

${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

S4.7.2 Multiply the number of pixels of the length by the square root of the real area per pixel obtained in the image calibration step to obtain the real distance from the cusp of each blade to the centroid, namely:

${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; the\; zi</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\sqrt{{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; dz</span>}}}$

The method of the present invention is further described below in conjunction with examples of implementation, and the method may further comprise the steps:

S1 In the planting pot of Arabidopsis thaliana, a calibration plate is placed, and RGB images of Arabidopsis thaliana are collected by a camera. The calibration board uses a black and white checkerboard with a blue frame, which consists of 3×3 squares with a side length of 4mm. Put it on the side of Arabidopsis thaliana, and perform image collection together with it, and the collected pictures are shown in Figure 3 (1).

S2 preprocesses the collected images to realize automatic correction and calibration of the images.

S2.1 Position black and white checkerboard

S2.1.1 According to the RGB characteristics of blue, the gray values of R and G are both smaller than B, and the gray value of B is greater than 150, and the blue border is segmented from the image.

S2.1.2 Fill the hole inside the image, and subtract the original blue border image from the result to obtain a new image.

S2.1.3 Perform an opening operation on the new image to remove noise points; then perform a closing operation to connect the breakpoints to obtain a black and white checkerboard area, as shown in Figure 3(2).

S2.2 Corner detection

S.2.2.1 Use the Prewitt template and image convolution to calculate the horizontal and vertical first-order derivatives of each point in the black and white checkerboard area, and obtain three new images: the square of the horizontal first-order derivative, the vertical one The square of the first derivative and the product of the two first derivatives.

S.2.2.2 Use a 101×101 Gaussian window to filter the three images to remove noise.

S.2.2.3 Compose the correlation matrix from the above three images, calculate the criterion function, and find out the corner points. As shown in Figure 3(3), each corner point has been marked with a red dot.

S2.3 Image correction and image calibration

S.2.3.1 The coordinates of the 16 square vertices of the checkerboard in the image from left to right and top to bottom obtained through corner point detection are as follows:

(163,370) (224,362) (286,354) (347,347)

(171,430) (233,423) (294,415) (356,407)

(179,491) (241,484) (302,476) (346,468)

(187,553) (249,545) (311,537) (372,529)

According to its strict square in the real world, the coordinates of each point are specified as:

(164,347) (225,347) (286,347) (347,347)

(164,408) (225,408) (286,408) (347,408)

(164,469) (225,469) (286,469) (347,469)

(164,530) (225,530) (286,530) (347,530)

Calculate the transformation matrix of these two sets of coordinates as:

$\left[\begin{array}{ccc}\mathrm{<span\; class="notranslate">\; 0.9820</span>}& \mathrm{<span\; class="notranslate">\; 0.1202</span>}& <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1.2519</span>}\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 05</span>}\\ <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0.1232</span>}& \mathrm{<span\; class="notranslate">\; 1.0126</span>}& \mathrm{<span\; class="notranslate">\; 3.8417</span>}\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 05</span>}\\ \mathrm{<span\; class="notranslate">\; 51.0888</span>}& <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 43.5533</span>}& \mathrm{<span\; class="notranslate">\; 0.9977</span>}\end{array}\right]$

S.2.3.2 The inverse of the transformation matrix is:

$\left[\begin{array}{ccc}\mathrm{<span\; class="notranslate">\; 1.0026</span>}& <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0.1183</span>}& \mathrm{<span\; class="notranslate">\; 0.0000</span>}\\ \mathrm{<span\; class="notranslate">\; 0.1237</span>}& \mathrm{<span\; class="notranslate">\; 0.9713</span>}& \mathrm{<span\; class="notranslate">\; 0.0000</span>}\\ <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 45.9428</span>}& \mathrm{<span\; class="notranslate">\; 48.4665</span>}& \mathrm{<span\; class="notranslate">\; 1.0000</span>}\end{array}\right]$

Apply it to the image to achieve image correction and restore the metric properties of the image. The result is shown in Figure 3(4).

S2.3.3 The total number of pixels of the black and white checkerboard is 35721 through the coordinates of the corner points, and according to its real size, that is, 9×(4mm) ² =144mm ² , the real size of the unit pixel is obtained, namely:

${\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; dz</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 144</span>}}{\mathrm{<span\; class="notranslate">\; 35721</span>}}$

S3 segments the preprocessed image, and extracts Arabidopsis from the image, specifically including the following sub-steps:

S3.1 Initial image segmentation

S3.1.1 Normalize the RGB value of each pixel to obtain rgb, extract the color difference map of 3g-2.4r-b, and use 0 as the threshold to binarize the image.

S3.1.2 Subtract the original detected blue frame area from the result. Determine whether the gray value of G of the pixel in the foreground area is greater than 50, if so, keep the pixel, otherwise determine the pixel as a background pixel and remove it from the area.

S3.1.3 Extract the area with the most pixels in the foreground connected area, which is the plant area, and the result is shown in Figure 4(1).

S3.2 Remove noise

S3.2.1 Use the 7×7 disk template to open the obtained image to obtain a new image, which will only retain the large leaf area and the center area of the plant, and remove the detailed parts of the image, including the stems and leaves of the plant of noise.

S3.2.2 Calculate the number of pixels in each connected region of the removed detail part, and remove the number of pixels less than 12 directly.

S3.2.3 The connected regions with the number of pixels greater than or equal to 12 are separately superimposed with the new image obtained by the open operation, and then calculate the number of regions in the superimposed image. If the number of regions decreases, it means that the connected region is a stem and must be retained , otherwise, if the number of regions in the image increases or remains unchanged after superimposition, it means that the connected region is the noise around the leaf and must be removed.

S3.2.4 After judging all the details, superimpose all the retained connected regions with the new image obtained by the open operation to obtain the final plant region, the result is shown in Figure 4(2).

S4. Extract the number of leaves of Arabidopsis thaliana, and calculate the distance from the tip of each leaf to the centroid, specifically including the following sub-steps:

S4.1 fills the holes in the plant area, and the result is shown in Figure 5(1).

S4.2 Through the contour search algorithm, extract the outer contour points of the filled plant area as shown in Fig. 5(2).

S4.3 Extract 1D signal

S4.3.1 After searching the contour, obtain the contour sequence, and extract the x-coordinate and y-coordinate of each contour point in the contour sequence.

S4.3.2 Calculate the distance from each outer contour point to the centroid of the plant area, as shown in Figure 5(3).

S4.4 Extended signal

The length of the 1D signal described in S4.4.1S4.3 is 1163, and the base 2 logarithm of 1163 is calculated, and the obtained result is rounded up to an integer and 1 is added to obtain the value 11.

S4.4.2 Because 2048 is less than 2326, the 1D signal described in S4.3 is mirrored and flipped, and after supplementing the 1D signal described in S4.3, the first 2048 data points are taken to obtain a new signal, as shown in the figure 5(4).

S4.5 Wavelet transform

S4.5.1 Perform 4-layer Haar wavelet decomposition on the 1D signal obtained in S4.4, and extract the wavelet coefficients of the 4th layer.

S4.5.2 Extract the first 1163 data points from the obtained wavelet coefficients to obtain a new 1D signal, making it the same length as the 1D signal described in S4.3.

S4.6 Find the positive zero-crossing point

S4.6.1 Shift the 1D signal S _C to the right and to the left by one bit respectively to obtain S _C1 and S _C2 .

S4.6.2 Find the positive zero-crossing point of this signal, and the result is shown in Figure 5(5). These positive zero-crossing points represent the local maximum of the 1D signal, as shown in Fig. 5(6). For a single leaf, its leaf cusp is farthest from the centroid of the plant area, ie the local maximum of the 1D signal is the cusp of each leaf. Therefore, the number of leaves of the plant can be obtained by finding the number of local maxima of the 1D signal.

S4.7 Blade length

S4.7.1 The coordinates of the tip of the blade in the image are:

(497,360) (485,232) (590,256) (607,324) (673,397) (600,480) (507,506 )(412,409 )

The coordinates of the centroid are: (542, 373)

Compute the distance from the cusp to the centroid of each leaf in the image:

d _i =[47.02 152.18 126.43 81.28 133.00 121.60 137.54 135.05]

S4.7.2 Multiply the number of pixels of the length by the square root of the real area per pixel obtained in the image calibration step to obtain the real distance from the cusp of each blade to the centroid, namely:

${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; the\; zi</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&CenterDot;</span>\sqrt{\frac{\mathrm{<span\; class="notranslate">\; 144</span>}}{\mathrm{<span\; class="notranslate">\; 35721</span>}}}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccccccc}\mathrm{<span\; class="notranslate">\; 3.0</span>}& \mathrm{<span\; class="notranslate">\; 9.7</span>}& \mathrm{<span\; class="notranslate">\; 8.0</span>}& \mathrm{<span\; class="notranslate">\; 5.2</span>}& \mathrm{<span\; class="notranslate">\; 8.4</span>}& \mathrm{<span\; class="notranslate">\; 7.7</span>}& \mathrm{<span\; class="notranslate">\; 8.7</span>}& \mathrm{<span\; class="notranslate">\; 8.6</span>}\end{array}\right]$

The unit is mm. The result is shown in Figure 5(7).

Claims (2)

1. Arabidopsis leaf number and blade cusp are to a detection method for the distance of barycenter, comprise the following steps:

S1., in the planting pot of arabidopsis, place scaling board, utilize the RGB image of collected by camera arabidopsis；

S2. the image after gathering is carried out pretreatment, it is achieved image from dynamic(al) correction and demarcation, wherein image calibration is precisely in order to school The distortion of positive image, image calibration is the full-size(d) in order to obtain unit picture element；

Step S2 includes positioning black and white gridiron pattern, Corner Detection, image rectification and image calibration；

Described location black and white gridiron pattern comprises the following steps: according to the blue component in RGB feature, extracts blue border；To figure The inside of picture carries out holes filling, then deducts original blue border image, obtains new images；New images is carried out opening operation, Remove noise；Carry out closed operation again, connect breakpoint, obtain the tessellated region of black and white；

Described Corner Detection comprises the following steps: calculate horizontal direction and Vertical Square that each in black and white gridiron pattern region is put First derivative upwards, obtains three width new images: horizontal first derivative square, two single orders of the quadratic sum of vertical first derivative The product of derivative；With gaussian filtering, three width images are filtered, remove noise；Correlation matrix is formed by above-mentioned three width images, Calculation criterion function, it is judged that whether pixel therein is angle point；

Described image rectification and image calibration comprise the following steps: obtain tessellated each foursquare top by Corner Detection Point coordinate in the picture, and according to its spatial relation in real world, obtain the transformation matrix of the two；Ask for conversion Inverse of a matrix, acts on image, it is achieved image rectification；The tessellated total number of pixels of black and white, and root is obtained by angular coordinate According to its full-size(d), obtain the full-size(d) of unit picture element；

S3. pretreated image is split, by arabidopsis and background segment, extract from image；

Step S3 includes image foremost segment and removes noise；

Described image foremost segment comprises the following steps: the rgb value of each pixel is normalized acquisition rgb, extracts 3g- The chromaticity difference diagram of 2.4r-b, carries out binaryzation as threshold value to image with 0；Result is deducted the blue border district originally detected Territory, it is judged that whether the gray value of the G of obtained foreground area pixel more than 50, the most then retains, otherwise remove；Extraction prospect is even The region with most pixel in logical region, is plant regional；

Described removal noise comprises the following steps: gained image is carried out opening operation, obtains new image, and it will only retain great Ye Panel region and the central area of plant, and remove the detail section in image, including the noise around the stem of plant and blade；Right Each connected region of the detail section removed calculates its number of pixels, number of pixels directly removing less than 12, number of pixels Connected region more than or equal to 12, the new images individually obtained with opening operation superposes, then the district calculated after superposition in image Territory number, if areal reduces, illustrates that this connected region is stem, it is necessary to retain, otherwise, if the district after superposition in image Territory number increases or constant, then illustrate that this connected region is the noise around blade, it is necessary to remove；Judge all of details After part, the new images that all connected regions retained all obtain with opening operation is superposed, obtains final plant regional；

S4. after being partitioned into arabidopsis image, extract arabidopsis blade number, and calculate each blade cusp to barycenter away from From；

Step S4 specifically includes following steps:

The hole of plant regional is filled with by S4.1, the barycenter of zoning；

S4.2 passes through profile lookup algorithm, extracts the outline point of the plant regional after filling；

S4.3 calculates each outline point distance to plant regional barycenter, thus 2 dimension (2D) images are become 1 1D signal；

S4.4 carries out data point mirror image filling to the 1D signal described in S4.3 so that it is expand to the integral number power times of 2, obtains new 1D signal；

The 1D signal that S4.4 is obtained by S4.5 carries out 4 layers of Haar wavelet decomposition, extract the 4th layer with the 1D Chief Signal Boatswain described in S4.3 Spend identical wavelet coefficient；

S4.6 searches the positive zero crossing in wavelet coefficient, the i.e. previous point of certain point, and less than 0, later point is more than 0, these What positive zero crossing represented is the local maximum of the 1D signal described in S4.3；For individual blade, its blade cusp away from Barycenter from plant regional is farthest, i.e. the local maximum of 1D signal is the cusp of each blade；Therefore by searching 1D signal Local maximum is achieved with the blade number of plant；

S4.7 calculates the cusp of each blade in image and, to the length of barycenter, the number of pixels of length is obtained with image calibration step To the actual length of unit picture element be multiplied, obtain the cusp actual distance to barycenter of each blade.

Arabidopsis leaf number the most according to claim 1 and blade cusp are to the detection method of the distance of barycenter, wherein Described scaling board uses rim black and white gridiron pattern, 3 × 3 length of sides be that the square of 4mm forms.