CN103729621B - Plant leaf image automatic recognition method based on leaf skeleton model - Google Patents

Abstract

The invention discloses a plant leaf image automatic recognition method based on a leaf skeleton model. The plant leaf image automatic recognition method based on the leaf skeleton model comprises the following steps of obtaining candidate points of a leaf skeleton in a plant leaf image, enhancing the image of the candidate points of the leaf skeleton, eliminating the noise in the plant leaf image, eliminating the leaf stem part, below the root bottom of each leaf, in the plant leaf image by color partition and smoothness partition, and obtaining the root bottom of each leaf and the main direction so as to recognize the position and the distribution direction of the leaves. The plant leaf image automatic recognition method based on the leaf skeleton model is high in processing efficiency and accuracy and wide in application range.

Description

Automatic recognition method of plant leaf image based on leaf skeleton model

technical field

The invention relates to the field of image processing research, in particular to a method for automatic recognition of plant leaf images based on a leaf skeleton model.

Background technique

Image segmentation is the subdivision of an image into objects that constitute its subregions. In the application, segmentation is stopped when the object of interest has been segmented. Image segmentation is the basis of image analysis and one of the most challenging problems in computer vision, which restricts the development and application of image processing and machine vision science. The images of plant leaves in the natural environment usually have the characteristics of similar background and plant color (for example, the background contains weeds, etc.), the leaves are easy to overlap, there are many leaves, and the boundary changes between leaves are not obvious. Due to these features, segmenting leaves in plant images is a very complex task.

In recent years, the research on segmentation and recognition of plant leaves has made some progress. Most of the research is mainly based on two categories of color and shape.

One, based on color:

Color-based segmentation methods can only segment images with large differences in leaf color and background color. Kristian Kirka et al. used the method of calculating the green and red bands in pixels to distinguish leaves from soil. Li Zhengming and others segmented the image based on the three orthogonal color feature quantities of R, G, and B component linear transformations. Zhu Weixing used it in the color space Y IQ, selected I as the feature quantity and used the improved maximum between-class variance method to separate plants and backgrounds. This type of segmentation method has great limitations. It can only segment the image with a large difference in color between the crop and the background into the crop part and the background part, and cannot separate the leaves in the image.

Second, based on shape:

Since the plants and their background (such as weeds in the background) are similar in color (mostly green) and similar in texture, it is relatively efficient to use shape-based segmentation. Wang, Z. et al. propose to use two stages to obtain the shape of the leaf. The first stage is to obtain the starting point of the blade: they search for the starting point of the blade model within a certain angle of the extension line of the skeleton, and select the point on the boundary within this angle that is farthest from the end point of the skeleton as the starting point of the blade model; the second stage is to describe the blade Type: first obtain the center point of the blade, calculate the distance to the center point for each corner point on the boundary, and obtain a sequence of distances to the center point. However, their main purpose is to classify the leaves. The segmentation algorithm is relatively simple. It only converts the color image into a grayscale image, uses the histogram to segment the leaves, and requires the leaves to be placed in a simple background. Carlos Caballero and M. Carmen Aranda describe leaf morphology by curvature based on points on the contour boundary. In the boundary point curvature histogram, select the extreme point of curvature (local maximum or minimum) as the feature point. The scale invariance is described by the ratio of the shortest distance between feature points to the total length of the boundary. Their main purpose is also to identify plant types. They also need to place the leaves in a simple background, and no segmentation algorithm is given. GuillaumeCerutti et al. proposed to use four integer parameters (top opening angle, bottom opening angle, relative maximum width, center position corresponding to the maximum width) and 10 points (three points on the bottom, three points on the top, two points on each side on both sides) points) describe the leaf polygonal model. For the leaves in the figure, a small area is first initialized with the model, and then the leaf boundaries are obtained by adjusting 14 parameters. But it is also based on a controllable environment, requiring the lens to face the leaf, the leaf is perpendicular to the abscissa, and the leaf is required to be in the center of the image, which is difficult to hold under natural conditions.

In the past two years, some scholars have also used three-dimensional technology to segment leaves. Long Quan et al. used a series of images surrounding the plants to reconstruct the three-dimensional state information of the plants, and separated the leaves with three-dimensional information and color information, but they needed dense images to cover the plants (generally 30-50), and required Manual interaction is difficult to apply on a large scale in practice. Chin-Hung Teng et al. used the optical flow field method to calculate the three-dimensional information of the leaves, and used the three-dimensional information to segment the leaves, but the leaves they dealt with were still relatively simple, and the results obtained by the method were prone to errors, and manual intervention was still required.

At present, the research on automatic recognition of complex leaf images based on natural environment is almost blank, which is mainly due to the complex structure of leaves in natural environment, messy distribution, and no obvious boundaries of leaf boundaries. At present, some important achievements are mostly in the research results of identifying plant species based on leaves. The main purpose of these studies is to obtain plant types through the study of the shape characteristics of plant leaves. The research methods are all based on human interaction, which cannot achieve the purpose of real-time automatic identification.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, and provide a method for automatic identification of plant leaf images based on a leaf skeleton model.

The purpose of the present invention is achieved through the following technical solutions:

A method for automatic identification of plant leaf images based on a leaf skeleton model, comprising steps in the following order:

1) Obtain the candidate points of the leaf skeleton in the plant leaf image:

a. Obtain the tangential direction of each pixel of the leaf, and assume that the number of pixels occupied by the width of the main stem of the leaf is d;

b. For each pixel point P, find two short lines whose two sides are parallel to the tangential direction, and then compare the pixel brightness of the point on the short line with the center point of the pixel point P, if the point whose brightness is smaller than the center point is above 65% , it is considered as a candidate point of the blade skeleton;

2) Enhance the image of the blade skeleton candidate point;

3) Eliminate the noise in the plant leaf image;

4) Eliminate the part of the leaf stem below the bottom of the leaf root in the plant leaf image:

a. Segmentation by color: keep the pixels whose green component is greater than the red component and blue component in the original image, and remove the interference of the non-green background;

b. Segmentation by smoothness: Calculate the smoothness of each pixel. When the smoothness of the pixel is greater than the threshold, the pixel is considered rough and does not belong to the leaf, and the pixel is removed;

c, obtain and remove the root image: the candidate point obtained in step 3) is checked with the image A through smoothness segmentation, if the candidate point is a pixel retained in the image A, then the candidate point is retained, otherwise it is deleted;

5) Obtain the bottom of each blade root and the main direction, thereby identifying the position and distribution direction of the blade:

a, step 4) the resulting figure obtained is refined, calculate the relative moments of inertia of all candidate points after refinement, get the point with the largest relative moments of inertia as the bottom of the blade root;

b. With the bottom of the blade root as the center, draw a rectangle at each angle α from 0 to 360 degrees. The direction of the two long sides of the rectangle is α, and the main direction of this rectangle is called α;

c. Count the number of candidate points in each rectangle, select the rectangle containing the most candidate points, the main direction α of this rectangle is the main direction of the blade, and the main direction of the blade refers to the direction from the bottom of the root to the top of the blade.

In step 1), described step a specifically includes the following steps:

S1. For each pixel point at each angle β of 0-360 degrees, draw a short line with the pixel point as the center, and the length of the short line is (3-6)×d;

S2. For each short line, calculate the brightness difference between other points of the short line and the central point, add up these differences, and select the short line direction with the smallest difference as the tangent direction of this pixel point.

Described step 2), specifically comprises the following steps:

S1. For each candidate point Q of the skeleton, draw a short line centered on the pixel point, the direction of the short line is the tangential direction of Q, the length of the short line is (2~3)×d, and check whether there are other candidates on the short line point, and its tangent direction is close to that of Q;

S2. If there is, set the intermediate point of the line connecting other detected candidate points and the candidate point Q as the candidate point, and use the tangent direction of the candidate point Q as the tangent direction of the intermediate point.

In step S1 of step 2), the criterion for judging that the tangential direction is similar is whether the angle difference between the two candidate points is within 5 degrees, and if so, the tangential direction of the two is considered to be similar.

Described step 3), specifically comprises the following steps:

S1. For each candidate point, take N points at each angle γ of 0-360 degrees, N=(4-7)×d, and judge the number of candidate points in the N points;

S2. If the proportion of the candidate points among the N points of one angle of the candidate point is greater than 65%, then keep the candidate point;

S3. If the proportion of the candidate points in each angle of 0-360 degrees of the candidate point is less than 65%, the candidate point is considered to be a noise point and deleted.

Step 4) in step b, the calculation formula of described smoothness is

${\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; D.</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; \&Psi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&Psi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>$

Among them, ψ(u,v) is the brightness of the candidate point P(u,v), D is a square area with P as the center and (6～10)×d+1 as the side length; ψ(i,j ) is the brightness value of other points in D except point P; M is the number of pixels in D.

In step a of step 5), the solution of the relative moment of inertia of the candidate point is specifically as follows:

S1. Assuming that the coordinates of the candidate point p are (u, v), and the coordinates of another candidate point q are (i, j), the angle between the line between the two candidate points and the tangent vector of the candidate point q represents is Angle(p,q), and the included angle is an acute angle;

S2. The relative moment of inertia of candidate point q relative to candidate point p can be divided into two cases:

If the angle between the line between the candidate point q and the candidate point p and the tangential direction of the candidate point p is greater than or equal to the threshold K, the relative moment of inertia of the candidate point q relative to the candidate point p is:

MoPtoP(p,q)=90-Angle(p,q);

If the angle between the line between the candidate point q and the candidate point p and the tangential direction of the candidate point p is smaller than the threshold K, the relative moment of inertia of the candidate point q relative to the candidate point p is:

MoPtoP(p,q)=Dis(p,q)×[90-Angle(p,q)];

Where Dis(p,q) is the distance between candidate point q and candidate point p;

Then the relative moment of inertia of the candidate point p is the sum of the relative moments of inertia of all other candidate points to the candidate point p:

RM _P =ΣMoPtoP(p,q).

In step b of step 1), the distance between the two short lines parallel to the tangential direction is (3-5)×d, and the length of each short line is (4-10)×d.

In step b of step 5), the rectangle has a length of (16-30)×d and a width of (7-12)×d.

Compared with the prior art, the present invention has the following advantages and beneficial effects:

1. High processing efficiency and high accuracy: Plant leaf images contain rich plant state information, which plays a vital role in plant state recognition. With the development of image acquisition technology, embedded technology, mobile technology and wireless transmission technology, people can collect a large number of plant leaf images anytime and anywhere, but at present, the analysis of these image plant leaves still mainly uses manual processing analysis: manual processing Low efficiency and poor timeliness cannot meet the requirements of real-time analysis. However, the method of the present invention automatically recognizes the skeleton of the blade in the plant image taken in a complex natural environment, and on this basis recognizes the bottom of the blade and the direction of the main stem from the bottom to the top of the blade, thereby automatically identifying the blade in the image. The leaf area provides premise preparation for plant state analysis, effectively shortens the processing time, improves processing efficiency, and can effectively avoid errors that occur during manual processing, with high accuracy.

2. Wide range of applications: the present invention can be widely used in fields such as botany and agricultural science. In the field of botany, the leaves of plants can be quickly identified, so as to further obtain the geometric characteristics of plant leaves, identify the types of plants, the change law of leaves during plant growth, and the distribution of leaves. In the field of agricultural science, the size and distribution of crop leaves can be automatically identified, and further analysis of the identified leaf parts can automatically identify the status of crops, such as water shortage, nutrient deficiency, and pest information, and provide early warning for farmers and management departments in a timely manner , thereby reducing agricultural losses.

Description of drawings

Fig. 1 is a flow chart of the method for automatic identification of plant leaf images based on a leaf skeleton model according to the present invention.

detailed description

The present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.

As shown in Figure 1, the method for automatic recognition of plant leaf images based on the leaf skeleton model includes the steps in the following order:

1) Obtain the candidate points of the leaf skeleton in the plant leaf image:

a. Obtain the tangential direction of each pixel of the leaf, and assume that the number of pixels occupied by the width of the main stem of the leaf is d, specifically including the following steps:

S1. For each pixel point at each angle β of 0-360 degrees, draw a short line with the pixel point as the center, and the length of the short line is (3-6)×d;

S2. For each short line, calculate the brightness difference between other points of the short line and the central point, add up these differences, and select the short line direction with the smallest difference as the tangential direction of this pixel point;

b. For each pixel point P, find two short lines whose two sides are parallel to the tangential direction, and then compare the pixel brightness of the point on the short line with the center point of the pixel point P, if the point whose brightness is smaller than the center point is above 65% , then it is considered as a candidate point of the blade skeleton; the distance between the two short lines parallel to the tangential direction is (3～5)×d, and the length of each short line is (4～10)×d;

2) Enhance the blade skeleton candidate point image, including the following steps:

S1. For each candidate point Q of the skeleton, draw a short line centered on the pixel point, the direction of the short line is the tangential direction of Q, the length of the short line is (2~3)×d, and check whether there are other candidates on the short line point, and its tangential direction is close to the tangential direction of Q; the criterion for judging the closeness of the tangential direction is whether the angle difference between the two candidate points is within 5 degrees, if so, the tangential direction of the two is considered to be close;

S2, if there is, then the intermediate point of the connection between other candidate points detected and the candidate point Q is also set as the candidate point, and the tangent direction of the candidate point Q is used as the tangent direction of the intermediate point;

3) Eliminate noise in the plant leaf image, specifically comprising the following steps:

S1. For each candidate point, take N points at each angle γ of 0-360 degrees, N=(4-7)×d, and judge the number of candidate points in the N points;

S2. If the proportion of the candidate points among the N points of one angle of the candidate point is greater than 65%, then keep the candidate point;

S3. If the ratio of the candidate points in each angle of 0 to 360 degrees of the candidate point is less than 65%, then the candidate point is considered to be a noise point and deleted;

4) Eliminate the part of the leaf stem below the bottom of the leaf root in the plant leaf image:

a. Segmentation by color: keep the pixels whose green component is greater than the red component and blue component in the original image, and remove the interference of the non-green background;

b. Segmentation by smoothness: calculate the smoothness of each pixel, and when the smoothness of the pixel is greater than the threshold, it is considered that the pixel is rough and does not belong to the blade, and the pixel is removed; the calculation formula of the smoothness for

${\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; D.</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; \&Psi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&Psi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>$

Among them, ψ(u,v) is the brightness of the candidate point P(u,v), D is a square area with P as the center and (6～10)×d+1 as the side length; ψ(i,j ) is the brightness value of other points in D except point P; M is the number of pixels in D; different types of leaves have different degrees of smoothness, and the threshold should be selected according to the type of leaf. Generally, the threshold is 40. If the leaf belongs to the rough type, the threshold Appropriate increase;

c, obtain and remove the root image: the candidate point obtained in step 3) is checked with the image A through smoothness segmentation, if the candidate point is a pixel retained in the image A, then the candidate point is retained, otherwise it is deleted;

5) Obtain the bottom of each blade root and the main direction, thereby identifying the position and distribution direction of the blade:

a, step 4) the resulting figure obtained is refined, calculate the relative moment of inertia of all candidate points after refinement, get the point with the largest relative moment of inertia as the bottom of the blade root; the solution of the relative moment of inertia of the candidate points is specifically as follows :

S1. Assuming that the coordinates of the candidate point p are (u, v), and the coordinates of another candidate point q are (i, j), the angle between the line between the two candidate points and the tangent vector of the candidate point q represents is Angle(p,q), and the included angle is an acute angle;

S2. The relative moment of inertia of candidate point q relative to candidate point p can be divided into two cases:

If the angle between the line between the candidate point q and the candidate point p and the tangent direction of the candidate point p is greater than or equal to the threshold K, the selection of the threshold is mainly to distinguish whether it is a pixel on the main stem or a pixel on the branch stem point, generally set to 18 degrees, then the relative moment of inertia of candidate point q relative to candidate point p is:

MoPtoP(p,q)=90-Angle(p,q);

If the angle between the line between the candidate point q and the candidate point p and the tangential direction of the candidate point p is smaller than the threshold K, the relative moment of inertia of the candidate point q relative to the candidate point p is:

MoPtoP(p,q)=Dis(p,q)×[90-Angle(p,q)];

Where Dis(p,q) is the distance between candidate point q and candidate point p;

Then the relative moment of inertia of the candidate point p is the sum of the relative moments of inertia of all other candidate points to the candidate point p:

RM _P = ΣMoPtoP(p, q);

b. Taking the bottom of the blade root as the center, draw a rectangle at each angle α of 0 to 360 degrees, the direction of the two long sides of the rectangle is α, and the main direction of this rectangle is called α; the length of the rectangle is (16 ~30)×d, the width is (7~12)×d;

c. Count the number of candidate points in each rectangle, select the rectangle containing the most candidate points, the main direction α of this rectangle is the main direction of the blade, and the main direction of the blade refers to the direction from the bottom of the root to the top of the blade.

The above-mentioned embodiment is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, Simplifications should be equivalent replacement methods, and all are included in the protection scope of the present invention.

Claims (9)

1. The plant blade image automatic recognition method based on blade skeleton model, comprises the step of following order: 1) Obtain the candidate points of the leaf skeleton in the plant leaf image: a. Obtain the tangential direction of each pixel of the leaf, and assume that the number of pixels occupied by the width of the main stem of the leaf is d; b. For each pixel point P, find two short lines whose two sides are parallel to the tangential direction, and then compare the pixel brightness of the point on the short line with the center point of the pixel point P, if the point whose brightness is smaller than the center point is above 65% , it is considered as a candidate point of the blade skeleton; 2) Enhance the image of the blade skeleton candidate point; 3) Eliminate the noise in the plant leaf image; 4) Eliminate the part of the leaf stem below the bottom of the leaf root in the plant leaf image: a. Segmentation by color: keep the pixels whose green component is greater than the red component and blue component in the original image, and remove the interference of the non-green background; b. Segmentation by smoothness: Calculate the smoothness of each pixel. When the smoothness of the pixel is greater than the threshold, the pixel is considered rough and does not belong to the leaf, and the pixel is removed; c, obtain and remove the root image: the candidate point obtained in step 3) is checked with the image A through smoothness segmentation, if the candidate point is a pixel retained in the image A, then the candidate point is retained, otherwise it is deleted; 5) Obtain the bottom of each blade root and the main direction, thereby identifying the position and distribution direction of the blade: a, step 4) the resulting figure obtained is refined, calculate the relative moments of inertia of all candidate points after refinement, get the point with the largest relative moments of inertia as the bottom of the blade root; b. With the bottom of the blade root as the center, draw a rectangle at each angle α from 0 to 360 degrees. The direction of the two long sides of the rectangle is α, and the main direction of this rectangle is called α; c. Count the number of candidate points in each rectangle, select the rectangle containing the most candidate points, the main direction α of this rectangle is the main direction of the blade, and the main direction of the blade refers to the direction from the bottom of the root to the top of the blade. 2. the plant blade image automatic recognition method based on blade skeleton model according to claim 1, is characterized in that: in step 1), described step a specifically comprises the following steps: S1. For each pixel point at each angle β of 0-360 degrees, draw a short line with the pixel point as the center, and the length of the short line is (3-6)×d; S2. For each short line, calculate the brightness difference between other points of the short line and the central point, add up these differences, and select the short line direction with the smallest difference as the tangent direction of this pixel point. 3. the plant blade image automatic recognition method based on blade skeleton model according to claim 1, is characterized in that: described step 2), specifically comprises the following steps: S1. For each candidate point Q of the skeleton, draw a short line centered on the pixel point, the direction of the short line is the tangential direction of Q, the length of the short line is (2~3)×d, and check whether there are other candidates on the short line point, and its tangent direction is close to that of Q; S2. If there is, set the intermediate point of the line connecting other detected candidate points and the candidate point Q as the candidate point, and use the tangent direction of the candidate point Q as the tangent direction of the intermediate point. 4. the plant blade image automatic recognition method based on blade skeleton model according to claim 3, it is characterized in that: in the step S1 of step 2), the judgment standard that described tangential direction is close is the angular difference of two candidate points Whether it is within 5 degrees, if yes, the tangent direction of the two is considered to be similar. 5. the plant blade image automatic recognition method based on blade skeleton model according to claim 1, is characterized in that: described step 3), specifically comprises the following steps: S1. For each candidate point, take N points at each angle γ of 0-360 degrees, N=(4-7)×d, and judge the number of candidate points in the N points; S2. If the proportion of the candidate points among the N points of one angle of the candidate point is greater than 65%, then keep the candidate point; S3. If the proportion of the candidate points in each angle of 0-360 degrees of the candidate point is less than 65%, the candidate point is considered to be a noise point and deleted. 6. the plant blade image automatic recognition method based on blade skeleton model according to claim 1, is characterized in that: in the step b of step 4), the computing formula of described smoothness is ${\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; D.</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; \&Psi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&Psi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>$ Among them, ψ(u,v) is the brightness of the candidate point P(u,v), D is a square area with P as the center and (6～10)×d+1 as the side length; ψ(i,j ) is the brightness value of other points in D except point P; M is the number of pixels in D. 7. the plant leaf image automatic recognition method based on blade skeleton model according to claim 1, is characterized in that: in the step a of step 5), the solution of described candidate point relative moment of inertia is specifically as follows: S1. Assuming that the coordinates of the candidate point p are (u, v), and the coordinates of another candidate point q are (i, j), the angle between the line between the two candidate points and the tangent vector of the candidate point q represents is Angle(p,q), and the included angle is an acute angle; S2. The relative moment of inertia of candidate point q relative to candidate point p can be divided into two cases: If the angle between the line between the candidate point q and the candidate point p and the tangential direction of the candidate point p is greater than or equal to the threshold K, the relative moment of inertia of the candidate point q relative to the candidate point p is: MoPtoP(p,q)=90-Angle(p,q); If the angle between the line between the candidate point q and the candidate point p and the tangential direction of the candidate point p is smaller than the threshold K, the relative moment of inertia of the candidate point q relative to the candidate point p is: MoPtoP(p,q)=Dis(p,q)×[90-Angle(p,q)]; Where Dis(p,q) is the distance between candidate point q and candidate point p; Then the relative moment of inertia of the candidate point p is the sum of the relative moments of inertia of all other candidate points to the candidate point p: RM _P =ΣMoPtoP(p,q). 8. the plant leaf image automatic recognition method based on blade skeleton model according to claim 1, it is characterized in that: in the step b of step 1), the distance between described two short lines parallel to the tangential direction is (3～ 5)×d, the length of each short line is (4～10)×d. 9. The plant leaf image automatic recognition method based on the leaf skeleton model according to claim 1, characterized in that: in step b of step 5), the length of the rectangle is (16～30) × d, and the width is ( 7～12)×d.