CN105046229A - Crop row identification method and apparatus - Google Patents

Abstract

The present invention discloses a crop row identification method and apparatus. The method comprises: firstly, converting a collected crop image to a binary image by using a fuzzy clustering method; secondly, extracting a region-of-interest (ROI) image from the obtained binary image, obtaining crop feature points by using a bar method, performing linear regression on the extracted crop feature points, and fitting out a crop row linear equation; and finally, removing wrong feature points by using the multiple linear regression method, and performing correction on the crop row linear equation. The apparatus comprises a fuzzy clustering module, a feature point extraction module and a fitting module. The method and the apparatus are high in crop identification accuracy, high in running speed and good in anti-interference capability.

Description

A kind of recognition methods of crop row and device

Technical field

The invention belongs to technical field of crop cultivation, particularly relate to a kind of recognition methods and device of the crop row based on image procossing.

Background technology

China is vast in territory, and landform, climate type complexity are various, are divided into the torrid zone, subtropics, temperate zone and frigid zone from south to north, and staple food crop has paddy rice, wheat, corn, soybean etc., and industrial crops have cotton, peanut, rape, sugarcane and beet etc.But China is populous, cultivated area is relatively less, and therefore particularly plant husbandry is extremely important in the status of China for agricultural, the whole national economy of relation.Along with mechanization of agriculture with informationalizedly to develop rapidly, the requirement realizing agricultural automation is more and more urgent.

Machine vision coordinates big-and-middle-sized agricultural machinery also more and more extensive in the utilization of agriculture field, and especially in vision guided navigation and crop identification, accuracy and expense are obtained for larger improvement.Therefore, in the work such as automatic weeding, results, fertilising, pruning, ploughing and weeding of crops, the identification carrying out crop row based on image procossing seems particularly important.

The recognition methods of existing crop row mainly utilizes super green method, maximum variance between clusters to split image, by Hough transform identification crop row.But these method Iamge Segmentation are inaccurate, can not well distinguish crops and weeds, and calculated amount is large, does not reach the requirement of real-time.Especially in crops, there are a large amount of weeds to deposit in case, desirable result can not be obtained.

Summary of the invention

The object of this invention is to provide a kind of recognition methods and device of crop row, to avoid prior art Iamge Segmentation inaccurate, the technical matters that recognition efficiency is not high.

To achieve these goals, technical solution of the present invention is as follows:

A recognition methods for crop row, described recognition methods comprises:

By fuzzy clustering method, the crop map picture of collection is converted into bianry image;

From the bianry image obtained, extract interested region ROI image, obtain wherein crops unique point by horizontal stripe method;

Linear regression is carried out to extracted crops unique point, simulates crop row straight-line equation;

Rejected the unique point of mistake by the method for repeatedly linear regression, crop row straight-line equation is revised.

Preferably, describedly by fuzzy clustering method, the crop map picture of collection is converted into bianry image, that the crop map picture that gathers is as whole sample, using the number percent shared by the value of pixel G passage as sample elements, the cluster centre of crops and background is initialized as respectively 0.35 ~ 0.40 and 0.30 ~ 0.35 and carries out fuzzy clustering and obtain.

The present invention, using crop map picture as whole sample, directly usually can carry out fuzzy clustering as the three-channel value of pixel R G B of pixel as sample using crop map.But preferably, adopt the number percent shared by value of pixel G passage as sample elements, thus reduce the dimension of cluster, the cluster centre of crops and background is initialized as respectively 0.35 ~ 0.40 and 0.30 ~ 0.35 and carries out fuzzy clustering and obtain bianry image, improve cluster speed.

Further, the wide of crop map picture of described collection is W pixel, and height is H pixel, and the wide of described ROI image is w=W/2 pixel, and wide is h=H/2, describedly obtains wherein crops unique point by horizontal stripe method, comprising:

ROI image is divided into the horizontal stripe that Q bar width is identical, uses S _p,qrepresent the number of times that in q article of horizontal stripe p row, white pixel occurs, wherein the value of p is the width pixel of ROI image from 1 to w, w;

For q article of horizontal stripe, to having threshold value u _q, threshold value u _qbe all S in q article of horizontal stripe _p,qaverage;

Work as S _p,qbe less than or equal to u _qand S _{p+1, q}be greater than u _qtime, think and enter crop row, record row coordinate is now p ₁;

Work as S _p,qbe more than or equal to u _qand S _{p+1, q}be less than u _qtime, think and leave crop row, record row coordinate is now p ₂;

Calculate the difference DELTA=p of columns when entering and leave crop row ₂-p ₁if Δ is greater than the constant d of setting, then think on horizontal stripe q from p ₁to p ₂section be crops, and to get this section of mid point be crops unique point;

Travel through all horizontal stripes, obtain crops unique points all in ROI image;

Wherein, the span of described constant d is: W/20<d<W/15.

Further, described linear regression is carried out to extracted crops unique point, simulates crop row straight-line equation, comprising:

According to the distribution of unique point, unique point is divided into different crop row;

For arbitrary crop row, if crop row straight-line equation is:

y＝kx+b

Wherein b is oblique distance, and k is slope, and calculating belongs to the distance l of all unique points to this straight line of this crop row:

$l=\frac{\left|kx+b-y\right|}{\sqrt{1+k^2}};$

Further calculating, all unique points belonging to this crop row are to the square distance of this straight line and l ':

$l^{\&prime;}=\underset{r=1}{\overset{M}{\&Sigma;}}{l_r}^2$

To k and b, local derviation asked to the molecule of above formula and make it be 0, obtaining:

$\frac{\&part;l^{\&prime;}}{\&part;b}=-2\underset{r=1}{\overset{M}{\&Sigma;}}(y_r-b-{\mathrm{kx}}_r)=0$

$\frac{\&part;l^{\&prime;}}{\&part;k}=-2\underset{r=1}{\overset{M}{\&Sigma;}}\&lsqb;y_r-(b+{\mathrm{kx}}_r)\&rsqb;x_r=0$

Wherein M is the quantity of all unique points belonging to this crop row, and r belongs to 1 ~ M, and the coordinate of r unique point is (x _r, y _r), l _rbe the distance of r unique point and crop row straight line, solve the solution of k and b in above formula and be respectively with carry it into straight-line equation to obtain crops straight-line equation and be: $y=\hat{k}x+\hat{b}.$

Further, the described method by repeatedly linear regression rejects the unique point of mistake, revises, comprising crop row straight-line equation:

According to crop row straight-line equation, calculating belongs to the distance of unique point to this crop row of this crop row, rejects the unique point that distance is greater than the constant of setting;

After the unique point rejecting mistake, again simulate crop row straight-line equation according to remaining unique point, and again according to the crop row straight-line equation newly simulated, calculate the distance of unique point to this straight line, reject the unique point that distance is greater than the constant of setting;

Circulation like this, until reach maximum linear return number of times, or when linear regression rejecting unique point quantity be 0 time iteration stopping.

The invention allows for a kind of recognition device of crop row, described device comprises:

Fuzzy clustering module, for being converted into bianry image by fuzzy clustering method by the crop map picture of collection;

Feature point extraction module, for extracting interested region ROI image from the bianry image obtained, obtains wherein crops unique point by horizontal stripe method;

Fitting module, for carrying out linear regression to extracted crops unique point, simulates crop row straight-line equation, and rejects the unique point of mistake by the method for repeatedly linear regression, revises crop row straight-line equation.

The recognition methods of a kind of crop row that the present invention proposes and device, by fuzzy clustering to collection Image Segmentation Using, utilize horizontal stripe method determination unique point, and utilize crop row belonging to the position judgment of unique point in x-axis, linear regression is utilized to obtain transition crop row, passing through the unique point of repeatedly linear regression rejecting mistake, draw final crop row.Identification crop row of the present invention accuracy is high, and travelling speed block, antijamming capability is strong.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of crop row recognition methods of the present invention;

Fig. 2 is embodiment of the present invention bianry image schematic diagram;

Fig. 3 is embodiment of the present invention ROI image schematic diagram;

Fig. 4 is unique point schematic diagram in embodiment of the present invention ROI image.

Embodiment

Be described in further details technical solution of the present invention below in conjunction with drawings and Examples, following examples do not form limitation of the invention.

The general thought of profit of the present invention is with computer vision technique, carries out analyzing and processing, identify crop row to the crop row image that image capture device collects.The present embodiment is described for corn seedling.

As shown in Figure 1, a kind of recognition methods of crop row, comprises the steps:

Step S1, by fuzzy clustering method, the crop map picture of collection is converted into bianry image.

The present embodiment X represents all samples, and the value of the RBG passage of each pixel of the crop map picture namely gathered is as a sample elements x _i, x _icorresponding to a pixel.Whole sample X is divided into crops and background two parts, so cluster numbers c=2, is divided into crops and background two class by the pixel in image.

Use f _ijrepresent x _ibelong to the degree of membership of jth class, use v _jrepresent the cluster centre of jth class, to each x in sample X _icarry out iteration, all sample elements have calculated and have been followed successively by iteration successively, are gone out the cluster centre of crops and background by iterative computation.

Need before the iteration cluster centre v _jcarry out initialization.Because crops part should be green in the ideal case, the present embodiment arranges crops cluster centre v ₁={ 0,255,0}; Remaining background Green passage, compared to red channel and blue channel, does not need to occupy an leading position, arranges v ₂={ 255,0,128}.

Degree of membership f _ijwith with cluster centre v _jiterative formula as follows:

$f_{ij}\left(t+1\right)=\frac{1}{{\mathrm{\&Sigma;}}_{r=1}^c{\&lsqb;\frac{d_{ij}\left(t\right)}{d_{ir}\left(t\right)}\&rsqb;}^{2/(\mathrm{\&lambda;}-1)}}$

$v_j(t+1)=\frac{{\mathrm{\&Sigma;}}_{r=1}^n{\&lsqb;f_{rj}\left(t\right)\&rsqb;}^{\mathrm{\&lambda;}}{x}_r}{{\mathrm{\&Sigma;}}_{r=1}^n{\&lsqb;f_{rj}\left(t\right)\&rsqb;}^{\mathrm{\&lambda;}}}$

Wherein, d _ijx _ito cluster centre v _jeuclidean distance, λ is referred to as index weight, and λ >1, n are the capacity of sample X.

This process iterate until || f _ij(t+1)-f _ij(t) || < ε or arrived the iterations t specified _max.

Two polymerization site: v will be obtained after iteration terminates ₁={ V _1R, V _1G, V _1Band v ₂={ V _2R, V _2G, V _2B, due to v ₁be the cluster centre of crops, the present embodiment is according to v ₁carry out computed segmentation threshold value, split preset=V _1G/ (V _1R+ V _1G+ V _1B).

For any pixel x _i, when the three-channel value of its RGB meet G/ (R+G+B) be greater than segmentation threshold time, this pixel is set to 255, and namely white, otherwise is set to 0, i.e. black, thus original image is divided into bianry image.Such as Fig. 2 is the secondary bianry image obtained according to collection image, and wherein white pixel represents crops, and black picture element represents background.

In different situations, crop information can well be extracted by cluster segmentation, compare and compare with the method for the combination of Da-Jin algorithm with conventional super green method, fuzzy clustering can preserve more details, such as weeds, fallen leaves etc., and under different illumination conditions, can accurately crops and background area be separated.

It should be noted that after cluster terminates, for each cluster centre V _j, the number percent val shared by the value of its G passage is:

$val=\frac{G}{R+B+G}$

By a large amount of experimental datas, as shown in table 1:

Background v 1(％)	Crop v 2(％)
		0.345	0.403
0.335	0.347
		0.345	0.386
0.336	0.353
		0.339	0.412
0.345	0.386

Table 1

Be not difficult to find that number percent val that crops cluster centres is corresponding is mostly between 0.35 and 0.40, number percent val corresponding to background cluster centre is between 0.30 and 0.35.And in actual cluster process, what play a decisive role is also the number percent shared by value of G passage.In order to improve counting yield, preferably, directly using the number percent shared by the value of G passage as sample elements, therefore the dimension of sample drops to 1 from 3, and before each cluster starts, the cluster centre of crops and background is initialized as 0.35 ~ 0.40 and 0.30 ~ 0.35 respectively, avoids double counting during each cluster iteration, improve cluster speed.

Step S2, from obtain bianry image extract interested region ROI, obtain wherein crops unique point by horizontal stripe method.

First, interested region ROI (RegionOfInterest) is extracted from the bianry image obtained, for wide be W pixel, height is the collection image of H pixel, the length of the ROI of general extraction is w=W/2, and wide is h=H/2, and the ROI that the present embodiment extracts as shown in Figure 3, the ROI extracted at least comprises a crop row, and the present embodiment comprises two crop row.

Horizontal stripe method is used to ROI, ROI image is divided into the horizontal stripe that Q bar width is identical, for wide be w pixel, height is the ROI image of h pixel, is divided into Q bar horizontal stripe, uses S _p,qrepresent the number of times that in q article of horizontal stripe p row, white pixel occurs, wherein the value of p is from 1 to w.

For q article of horizontal stripe, to having threshold value u _q, threshold value u _qbe all S in q article of horizontal stripe _p,qaverage: $u_q=\frac{1}{w}{\&Sigma;}_{p=0}^{p=w}{S}_{p,q}.$

Traversal S _{p, q}, the unique point with following procedure extraction crop row:

(1) S is worked as _p,qbe less than or equal to u _qand S _{p+1, q}be greater than u _qtime, illustrate and enter crop row, record row coordinate is now p ₁;

(2) S is worked as _p,qbe more than or equal to u _qand S _{p+1, q}be less than u _qtime, illustrate and leave crop row, record row coordinate is now p ₂;

(3) enter at every turn and leave crop row time, calculate the difference DELTA=p of columns when entering and leave crop row ₂-p ₁if Δ is greater than the constant d of setting, then think on horizontal stripe q from p ₁to p ₂section be crops, and to get this section of mid point be crops unique point.

Wherein, d is constant, can be understood as the width of crop row, and its span is: W/20<d<W/15.

By traveling through all horizontal stripes, can obtain some unique points, as shown in Figure 4, wherein stain is unique point.

Step S3, linear regression is carried out to extracted crops unique point, draw crop row straight-line equation.

According to the distribution of unique point, unique point is divided into different crop row.In the ROI that the present embodiment creates, general only reservation two crop row, judge the x-axis coordinate of all unique points, if coordinate is less than the half of width, this point are attributed to the crop row in left side, otherwise are attributed to right side.

For arbitrary crop row, all unique points according to belonging to this crop row obtain crop row straight-line equation by linear regression, and process is as follows:

Suppose that the coordinate of unique point is for (x, y), the equation of crop row straight line is:

y＝kx+b

Wherein, b is oblique distance, and k is slope.

Then unique point to the distance l of this crop row straight line is:

$l=\frac{\left|kx+b-y\right|}{\sqrt{1+k^2}};$

The quadratic sum l ' of all feature distance between beeline and dot is:

$l^{\&prime;}=\underset{r=1}{\overset{M}{\&Sigma;}}{l_r}^2;$

Wherein M is the quantity of unique point.

Visible, if unique point (x, y) on this line, then l is 0, but can not all unique points all just fall on this line, by asking k and b to make l ' get minimum value, namely right k and b of molecule ask local derviation and make it be 0:

$\frac{\&part;l^{\&prime;}}{\&part;b}=-2\underset{r=1}{\overset{M}{\&Sigma;}}(y_r-b-{\mathrm{kx}}_r)=0$

$\frac{\&part;l^{\&prime;}}{\&part;k}=-2\underset{r=1}{\overset{M}{\&Sigma;}}\&lsqb;y_r-(b+{\mathrm{kx}}_r)\&rsqb;x_r=0$

Wherein M is the quantity of all unique points belonging to this crop row, and r belongs to 1 ~ M, and the coordinate of r unique point is (x _r, y _r), l _rbe the distance of r unique point and crop row straight line, solve the solution of above formula with will with bring straight-line equation into, the equation obtaining crop row straight line is:

$y=\hat{k}x+\hat{b}.$

Unique point is distributed near these straight line both sides substantially.

Step S4, rejected the unique point of mistake by the method for repeatedly linear regression, crop row straight-line equation is revised.

Ideally, crop row straight-line equation now by drawing in step S3 can obtain crop row comparatively accurately, but only may not comprise two crop row in ROI in actual conditions, and illumination, weather, the extraneous factors such as weeds also can affect accuracy.In order to obtain crop row accurately, the present embodiment needs the unique point rejecting mistake, retains correct unique point.

In all external environments, the intrusion of other row and having the greatest impact of weeds in ROI, and these crop row invaded and weeds are spectrally close with target, so the present embodiment utilizes point to judge which unique point needs to reject with the position relationship of line.

The present embodiment introduces the method for repeatedly linear regression, by S3, has obtained crop row straight-line equation and by the distance l of distance between beeline and dot formulae discovery unique point to this straight line, when l is greater than W/15, reject this point, W is the wide of image.

After the unique point rejecting mistake, again simulate the straight-line equation of crop row according to remaining unique point, and again according to the straight-line equation of the crop row newly simulated, calculate unique point to the distance l of this straight line, when l is greater than W/15, reject this point.

Circulation like this, until reach maximum linear return number of times, the present embodiment is 10 times, or when linear regression rejecting unique point quantity be 0 time iteration stopping.

The present embodiment, in order to raise the efficiency, separately carries out linear regression to two crop row, when the unique point on a crop row does not have mistake, directly skips and solves another crop row.

The present embodiment also proposed the recognition device of a kind of crop row corresponding to said method, and described device comprises:

Fuzzy clustering module, for being converted into bianry image by fuzzy clustering method by the crop map picture of collection;

Feature point extraction module, for extracting interested region ROI image from the bianry image obtained, obtains wherein crops unique point by horizontal stripe method;

Fitting module, for carrying out linear regression to extracted crops unique point, simulates crop row straight-line equation, and rejects the unique point of mistake by the method for repeatedly linear regression, revises crop row straight-line equation.

Operation performed by each module is corresponding with above-mentioned crop row recognition methods, repeats no more here.

The contrast of the repeatedly linear regression that prior art adopts Hough transform and the present embodiment to propose is as shown in table 2:

Table 2

Numerical value in table 2 is mean values.As can be seen from the table, be no matter that the latter is all better than the former, and when image increases gradually, the calculated amount increase tendency of Hough transform is also faster than the method for the present embodiment in accuracy or consuming time.Experimental data display after testing corn seedling, under the condition of different weather environment, identify that the rate of accuracy reached of crop row is to 96%, error is at about 2 °, and travelling speed, within 10ms, can reach the requirement of real-time.

Above embodiment is only in order to illustrate technical scheme of the present invention but not to be limited; when not deviating from the present invention's spirit and essence thereof; those of ordinary skill in the art are when making various corresponding change and distortion according to the present invention, but these change accordingly and are out of shape the protection domain that all should belong to the claim appended by the present invention.

Claims (10)

1. a recognition methods for crop row, is characterized in that, described recognition methods comprises:

By fuzzy clustering method, the crop map picture of collection is converted into bianry image;

From the bianry image obtained, extract interested region ROI image, obtain wherein crops unique point by horizontal stripe method;

Linear regression is carried out to extracted crops unique point, simulates crop row straight-line equation;

Rejected the unique point of mistake by the method for repeatedly linear regression, crop row straight-line equation is revised.

2. the recognition methods of crop row according to claim 1, it is characterized in that, describedly by fuzzy clustering method, the crop map picture of collection is converted into bianry image, that the crop map picture that gathers is as whole sample, using the number percent shared by the value of pixel G passage as sample elements, the cluster centre of crops and background is initialized as respectively 0.35 ~ 0.40 and 0.30 ~ 0.35 and carries out fuzzy clustering and obtain.

3. the recognition methods of crop row according to claim 1, is characterized in that, the wide of crop map picture of described collection is W pixel, height is H pixel, and the wide of described ROI image is w=W/2 pixel, and wide is h=H/2, describedly obtain wherein crops unique point by horizontal stripe method, comprising:

ROI image is divided into the horizontal stripe that Q bar width is identical, uses S _p,qrepresent the number of times that in q article of horizontal stripe p row, white pixel occurs, wherein the value of p is the width pixel of ROI image from 1 to w, w;

For q article of horizontal stripe, to having threshold value u _q, threshold value u _qbe all S in q article of horizontal stripe _p,qaverage;

Work as S _p,qbe less than or equal to u _qand S _{p+1, q}be greater than u _qtime, think and enter crop row, record row coordinate is now p ₁;

Work as S _p,qbe more than or equal to u _qand S _{p+1, q}be less than u _qtime, think and leave crop row, record row coordinate is now p ₂;

Calculate the difference DELTA=p of columns when entering and leave crop row ₂-p ₁if Δ is greater than the constant d of setting, then think on horizontal stripe q from p ₁to p ₂section be crops, and to get this section of mid point be crops unique point;

Travel through all horizontal stripes, obtain crops unique points all in ROI image;

Wherein, the span of described constant d is: W/20<d<W/15.

4. the recognition methods of crop row according to claim 1, is characterized in that, describedly carries out linear regression to extracted crops unique point, simulates crop row straight-line equation, comprising:

According to the distribution of unique point, unique point is divided into different crop row;

For arbitrary crop row, if crop row straight-line equation is:

y＝kx+b

Wherein b is oblique distance, and k is slope, and calculating belongs to the distance l of all unique points to this straight line of this crop row:

$l=\frac{\left|kx+b-y\right|}{\sqrt{1+k^2}};$

Further calculating, all unique points belonging to this crop row are to the square distance of this straight line and l ':

$l^{\&prime;}=\underset{r=1}{\overset{M}{\&Sigma;}}{l_r}^2$

To k and b, local derviation asked to the molecule of above formula and make it be 0, obtaining:

$\frac{\&part;l^{\&prime;}}{\&part;b}=-2\underset{r=1}{\overset{M}{\&Sigma;}}(y_r-b-{\mathrm{kx}}_r)=0$

$\frac{\&part;l^{\&prime;}}{\&part;k}=-2\underset{r=1}{\overset{M}{\&Sigma;}}\&lsqb;y_r-(b+{\mathrm{kx}}_r)\&rsqb;x_r=0$

Wherein M is the quantity of all unique points belonging to this crop row, and r belongs to 1 ~ M, and the coordinate of r unique point is (x _r, y _r), l _rbe the distance of r unique point and crop row straight line, solve the solution of k and b in above formula and be respectively with carry it into straight-line equation to obtain crops straight-line equation and be: $y=\hat{k}x+\hat{b}.$

5. the recognition methods of crop row according to claim 4, is characterized in that, the described method by repeatedly linear regression rejects the unique point of mistake, revises, comprising crop row straight-line equation:

According to crop row straight-line equation, calculating belongs to the distance of unique point to this crop row of this crop row, rejects the unique point that distance is greater than the constant of setting;

After the unique point rejecting mistake, again simulate crop row straight-line equation according to remaining unique point, and again according to the crop row straight-line equation newly simulated, calculate the distance of unique point to this straight line, reject the unique point that distance is greater than the constant of setting;

Circulation like this, until reach maximum linear return number of times, or when linear regression rejecting unique point quantity be 0 time iteration stopping.

6. a recognition device for crop row, is characterized in that, described device comprises:

Fuzzy clustering module, for being converted into bianry image by fuzzy clustering method by the crop map picture of collection;

Feature point extraction module, for extracting interested region ROI image from the bianry image obtained, obtains wherein crops unique point by horizontal stripe method;

Fitting module, for carrying out linear regression to extracted crops unique point, simulates crop row straight-line equation, and rejects the unique point of mistake by the method for repeatedly linear regression, revises crop row straight-line equation.

7. the recognition device of crop row according to claim 6, it is characterized in that, described fuzzy clustering module is when being converted into bianry image by fuzzy clustering method by the crop map picture of collection, that the crop map picture that gathers is as whole sample, using the number percent shared by the value of pixel G passage as sample elements, the cluster centre of crops and background is initialized as respectively 0.35 ~ 0.40 and 0.30 ~ 0.35 and carries out fuzzy clustering and obtain.

8. the recognition device of crop row according to claim 6, it is characterized in that, the wide of crop map picture of described collection is W pixel, height is H pixel, the wide of described ROI image is w=W/2 pixel, wide is h=H/2, described feature point extraction module by horizontal stripe method obtain wherein crops unique point time, perform as follows operate:

ROI image is divided into the horizontal stripe that Q bar width is identical, uses S _p,qrepresent the number of times that in q article of horizontal stripe p row, white pixel occurs, wherein the value of p is the width pixel of ROI image from 1 to w, w;

For q article of horizontal stripe, to having threshold value u _q, threshold value u _qbe all S in q article of horizontal stripe _p,qaverage;

Work as S _p,qbe less than or equal to u _qand S _{p+1, q}be greater than u _qtime, think and enter crop row, record row coordinate is now p ₁;

Work as S _p,qbe more than or equal to u _qand S _{p+1, q}be less than u _qtime, think and leave crop row, record row coordinate is now p ₂;

Calculate the difference DELTA=p of columns when entering and leave crop row ₂-p ₁if Δ is greater than the constant d of setting, then think on horizontal stripe q from p ₁to p ₂section be crops, and to get this section of mid point be crops unique point;

Travel through all horizontal stripes, obtain crops unique points all in ROI image;

Wherein, the span of described constant d is: W/20<d<W/15.

9. the recognition device of crop row according to claim 6, is characterized in that, described fitting module carries out linear regression to extracted crops unique point, when simulating crop row straight-line equation, performs and operates as follows:

According to the distribution of unique point, unique point is divided into different crop row;

For arbitrary crop row, if crop row straight-line equation is:

y＝kx+b

Wherein b is oblique distance, and k is slope, and calculating belongs to the distance l of all unique points to this straight line of this crop row:

$l=\frac{\left|kx+b-y\right|}{\sqrt{1+k^2}};$

Further calculating, all unique points belonging to this crop row are to the square distance of this straight line and l ':

$l^{\&prime;}=\underset{r=1}{\overset{M}{\&Sigma;}}{l_r}^2$

To k and b, local derviation asked to the molecule of above formula and make it be 0, obtaining:

$\frac{\&part;l^{\&prime;}}{\&part;b}=-2\underset{r=1}{\overset{M}{\&Sigma;}}(y_r-b-{\mathrm{kx}}_r)=0$

$\frac{\&part;l^{\&prime;}}{\&part;k}=-2\underset{r=1}{\overset{M}{\&Sigma;}}\&lsqb;y_r-(b+{\mathrm{kx}}_r)\&rsqb;x_r=0$

Wherein M is the quantity of all unique points belonging to this crop row, and r belongs to 1 ~ M, and the coordinate of r unique point is (x _r, y _r), l _rbe the distance of r unique point and crop row straight line, solve the solution of k and b in above formula and be respectively with carry it into straight-line equation to obtain crops straight-line equation and be: $y=\hat{k}x+\hat{b}.$

10. the recognition device of crop row according to claim 9, is characterized in that, described fitting module rejects the unique point of mistake in the method by repeatedly linear regression, when revising crop row straight-line equation, performs and operates as follows:

According to crop row straight-line equation, calculating belongs to the distance of unique point to this crop row of this crop row, rejects the unique point that distance is greater than the constant of setting;

After the unique point rejecting mistake, again simulate crop row straight-line equation according to remaining unique point, and again according to the crop row straight-line equation newly simulated, calculate the distance of unique point to this straight line, reject the unique point that distance is greater than the constant of setting;

Circulation like this, until reach maximum linear return number of times, or when linear regression rejecting unique point quantity be 0 time iteration stopping.