CN105654099A - Sugarcane segmentation and identification method based on improved vision - Google Patents

Abstract

To monitor the growth of sugarcanes or intelligently cut off seed sugarcanes containing sugarcane shoots, a method for automatically completing sugarcane segmentation and identification by identifying the shape and stem node characteristics of sugarcanes based on an improved computer vision technology is put forward. First, sugarcane images are obtained by a digital device; then, hue saturation intensity (HIS) color space conversion is performed on the sugarcane images, color characteristics and threshold segmentation of H and S components are combined, AND operation is performed on reverse images after threshold segmentation to get a composite image, and the composite image is divided into 64 column areas; and finally, the characteristic indexes of H parameters, S parameters, roughness ratio, white spot ratio and the like are extracted, and nodes and inter-node columns are classified and identified through a support vector machine so as to complete identification of the nodes and positions, and the average identification rate is 94.2%.

Description

Based on the Caulis Sacchari sinensis segmentation and the recognition methods that improve vision

Technical field

The invention belongs to technical field of image processing, be a kind of Caulis Sacchari sinensis feature is extracted and knows method for distinguishing in conjunction with Caulis Sacchari sinensis image and computation model.

Background technology

In the growth and end processing sequences of Caulis Sacchari sinensis, the cutting of growth conditions and Caulis Sacchari sinensis sugarcane bud is manually carried out for a long time, this method passes through identification and the process of computer vision technique, the image of Caulis Sacchari sinensis can be identified automatically, need the Caulis Sacchari sinensis by whole to cut into the effective sugarcane kind fragment comprising 1��3 sugarcane bud in sugarcane kind processes. Currently also be manually performed is main more. For improving efficiency, reduce labor intensity and realize becoming more meticulous of cane planting, it is necessary to developing the intelligent shearing device of recognizable stipes and internode, and wherein it is crucial that identify cane stalk. The domestic research in this field at present still belongs to blank. Close research has Liu Qingting etc. to utilize High-speed Photography Analysis blade cutting sugarcane genotypes destructive process, and abroad, only Iran MoshashaiK utilizes the method for gray level image Threshold segmentation that cane stalk identification has been done preliminary study, also also in the starting stage. There is not been reported in the research of domestic Caulis Sacchari sinensis stem image automatic identification technology. The present invention adopts the Segmentation of Color Image based on color feature vector that Caulis Sacchari sinensis image is split, then the stipes class identified is carried out clustering recognition, it is thus achieved that stipes number and stipes position, achieves good effect.

Summary of the invention

Based on the Caulis Sacchari sinensis segmentation and the recognition methods that improve vision, it is characterized in that comprising step in detail below:

(1) Caulis Sacchari sinensis original image is obtained;

(2) it is transformed into HSI color space by original image denoising with from RGB color, and H in Selection Model and S parameter are as the feature of Caulis Sacchari sinensis image;

(3) Caulis Sacchari sinensis image is carried out the process of convolution;

(4) Caulis Sacchari sinensis image is carried out the extraction of feature;

(5) Caulis Sacchari sinensis characteristics of image carries out model calculate and mate;

(6) the Caulis Sacchari sinensis image completing feature extraction is carried out SVM identification.

Concrete process includes:

1, the acquisition of image

Adopting the single Caulis Sacchari sinensis coloured image of CANONS80 type digital camera shooting red background on testing stand, image is sized to 1600*1200 pixel, JPG form. Process software and adopt VC++6.0 and Matlab7.0. First leaf of Caulis Sacchari sinensis is peeled off before shooting, camera lens and table vertical, distance workbench 30cm.

2, image basic handling

Caulis Sacchari sinensis surface color transition from black to white in image, occupies whole gray level, adopts RGB color to be difficult to be partitioned into desirable Caulis Sacchari sinensis profile, and the dependency of each component space is also difficult to embody stipes and internode feature difference. After substantial amounts of Experimental Comparison, it has been found that adopt the HSI space of red background effectively background to be distinguished with Caulis Sacchari sinensis.

(1) based on the color space of HSI

HSI space is relatively more directly perceived and meets human vision property, and H, S, I represent tone, saturation and brightness. From the H component map of HSI color space it can be seen that the gray scale of Caulis Sacchari sinensis and background area is spatially disperseed, the tone of stipes and internode has notable difference, can as the basis of characterization of stipes Yu internode; It can be seen that Caulis Sacchari sinensis is clear-cut from S component, background area uniform gray level, its grey level histogram manifests bimodality, is conducive to Caulis Sacchari sinensis contours extract.

(2) process of convolution

After the process of convolution Color Segmentation of bianry image, image is bianry image. By amplifying observation it can be seen that the image after segmentation be made up of the region of substantial amounts of, independent point and little area, it is impossible to carry out image further effectively splitting. Therefore, process of convolution is adopted, it is simple to follow-up image processing process. Process of convolution is a kind of linearity filtering method. The investigation and comparison mask of 7 kinds of types, finally determines the DISK mask process of application 5 �� 5 by experiment.

(3) Threshold segmentation

H component can embody the minutia of cane stalk, and S component embodies the contour feature of Caulis Sacchari sinensis. Select suitable Threshold segmentation can obtain desirable Caulis Sacchari sinensis profile and characterize the binary map of cane stalk and internode difference; On S, H threshold selection method, respectively Otsu is done comparative test with artificial selection's threshold value, it has been found that Otsu is split automatically can not effectively eliminate background noise, there is large-area reflective phenomenon. And when workbench control, the rectangular histogram trough of S component occurs between 0.40��0.55, threshold value is set between and can obtain desirable Caulis Sacchari sinensis boundary profile figure; Adding up gray level with 0.1 interval region in 0.4��0.6 scope to count, in 0.1 scope, minimum gray level counts the gray value of correspondence as the segmentation threshold of S component. Research finds that in H histogram of component, the pixel of the overwhelming majority is between gray level 0��0.04 and 0.90��1, arranges threshold value, it is possible to effectively obtain stipes feature between interval 0.85��0.15.

3, feature extraction

Owing to the white point number of eustipes part disperses relatively uniform, the white point number in stipes region is intensive, and diameter is relatively larger; Be 64 row block regions by composite diagram divided by column, if image collection be X (i, j); The top edge of image is P_t=(x_t, y_t), lower limb is P_t=(x_t, y_t), kth row block rugosity is

$C_k=\frac{1}{10}\underset{i=10(k+1)}{\overset{10k}{\mathrm{\Σ}}}(y_{\mathrm{bi}}-y_{\mathrm{ti}})$

The rugosity (i.e. diameter) of each row block is rugosity ratio with the ratio of maximum rugosity

$S_k=\frac{C_k}{\max \left(C_k\right)}$

With lower boundary, each row block being divided 8 by the coboundary of each row block and wait row block, the white point number sum of center 4 row block and the ratio of the white point sum of its column are district's ratio in 1/2, and its computing formula is

$I_k=\frac{{\mathrm{\Σ}}_{f=3}^6{R}_{\mathrm{kf}}}{W_k}$

Each row block rugosity and both sides 5 row block distance two row block rugosity around and average ratio be position slightly than, its computing formula is

$Y_k=\left\{\begin{array}{cc}\frac{{2C}_k}{C_{k-5}+C_{k+5}}& (k\∈[\mathrm{6,59}\left]\right)\\ \frac{{2C}_k}{C_1+C_{k+5}}& (k\∈[\mathrm{1,5}\left]\right)\\ \frac{{2C}_k}{C_{k-5}+C_{64}}& (k\∈[\mathrm{60,64}\left]\right)\end{array}\right.$

4, SVM recognizer identification stipes and internode row block

SVM is a kind of new mode identification method, it takes into account training error and generalization ability, many distinctive advantages are shown in solving small sample, dimension non-linear, high, local minimum isotype identification problem, in two quasi-mode identification problems, given training data { (x_i, y_i);X_i��R^N; y_i=�� 1}, supports the classifying rules that vector is determined by f (x)

$f\left(x\right)=\mathrm{sgn}[\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{a}_{i}K(x_i,x)+b]$

The optimal solution of the optimization problem that application Lagrange multiplier method obtains

$\max \underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i-\frac{1}{2}\underset{i,j=1}{\overset{l}{\mathrm{\Σ}}}{a}_i{a}_j{y}_i{y}_{j}K(x_i,x_j),0\≤a_i\≤C$

5, clustering recognition stipes number and position

The row block distribution obtained from svm classifier: the stipes row block identified is not unique, this and stipes have certain width to be consistent, the purpose of cane stalk identification is in that the cutting off tool controlled to controller finds cutting position, expresses by cane stalk joint number discrimination in picture and stipes location recognition rate. Stipes number that stipes number discrimination is defined as from image to identify through algorithm and the percentage ratio of actual stipes number in figure. For characterizing stipes location recognition rate, introducing and cut just rate concept, just rate of cutting when falling into internode center with tool position is expressed as 100%, and just rate cut by cutter when falling into stipes center be 0. The method adopting cluster can find stipes region. Finally adopt knearest neighbour method to stipes class cluster analysis.

6, beeline cluster

Knearest neighbour method rule is: as long as the minimum range of two classes is less than threshold value, and two classes are just merged into a class, defines D_{I, j}For ��_iMinimum range between all samples of apoplexy due to endogenous wind, i.e. D_ij=min{d_UV}

Wherein d_UVFor ��_iApoplexy due to endogenous wind sample U class and ��_jDistance between apoplexy due to endogenous wind sample V. If ��_jClass is by ��_m����_nTwo classes merging form, then

D_im=min{d_UA}(U�ʦ�_iClass, A �� ��_mClass)

D_im=min{d_UB}(U�ʦ�_iClass, B �� ��_nClass)

Accompanying drawing explanation

Fig. 1 is Caulis Sacchari sinensis Feature extraction and recognition method system

Fig. 2 is the composograph of Caulis Sacchari sinensis

Detailed description of the invention

50 width Caulis Sacchari sinensis picture combined training storehouses are extracted for checking from the image gathered. Process through primary image, extract 50 width images, every 64 row block, totally 3200 samples, calculate each sample characteristics index; Through the method for artificial cognition, divide the category attribute of 3200 samples. Statistics having found that, the row block ratio of internode class and stipes class reaches 10: 1 in piece image, need to extract the training sample that ratio is suitable between class and carry out training pattern, thus extract again from sample whole stipes class samples and part internode class sample totally 800 set up disaggregated model. SVM arranges C=20, G=0.01 through cross matching.

What SVM identified realizes step:

(1) obtain the SVM stipes class identified, calculate the quantity Nm of stipes class row block, with the positional distance between stipes class row block for characteristic parameter.

(2) the minimum threshold of distance T arranging cluster is 20��30 (row block distances).

(3) by all each point of classes of stipes row block, cluster centre number is Nm.

(4) all stipes row blocks are circulated, find two nearest row block pi, pj, if distance is D; If D is less than or equal to T, merging pi, pj, big for class-mark is included into the apoplexy due to endogenous wind that class-mark is little, otherwise, D, more than T, exits circulation. The class number obtained is stipes number.

Claims (4)

1., based on the Caulis Sacchari sinensis segmentation and the recognition methods that improve vision, it is characterized in that comprising step in detail below:

(1) Caulis Sacchari sinensis original image is obtained;

(2) it is transformed into HSI color space by original image denoising with from RGB color, and H in Selection Model and S parameter are as the feature of Caulis Sacchari sinensis image;

(3) Caulis Sacchari sinensis image is carried out the process of convolution;

(4) Caulis Sacchari sinensis image is carried out the extraction of feature;

(5) Caulis Sacchari sinensis characteristics of image carries out model calculate and mate;

(6) the Caulis Sacchari sinensis image completing feature extraction is carried out SVM identification.

2. according to claim 1 based on the Caulis Sacchari sinensis segmentation and the recognition methods that improve vision, it is characterized in that:

Image be have employed process of convolution, in order to follow-up image processing process.Process of convolution is a kind of linearity filtering method. The investigation and comparison mask of 7 kinds of types, finally determines the DISK mask process of application 5 �� 5 by experiment.

3. according to claim 1 based on the Caulis Sacchari sinensis segmentation and the recognition methods that improve vision, it is characterized in that:

In the extraction of feature, be 64 row block regions by composite diagram divided by column, if image collection be X (i, j); The top edge of image is P_t=(x_t, y_t), lower limb is P_t=(x_t, y_t), kth row block rugosity is

The rugosity (i.e. diameter) of each row block is rugosity ratio with the ratio of maximum rugosity

With lower boundary, each row block being divided 8 by the coboundary of each row block and wait row block, the white point number sum of center 4 row block and the ratio of the white point sum of its column are district's ratio in 1/2, and its computing formula is

Each row block rugosity and both sides 5 row block distance two row block rugosity around and average ratio be position slightly than, its computing formula is

��

4. according to claim 1 based on the Caulis Sacchari sinensis segmentation and the recognition methods that improve vision, it is characterized in that:

In SVM image recognition

In pattern recognition problem, given training data { (x_i, y_i); x_i��R^N; y_i=�� 1}, supports the classifying rules that vector is determined by f (x)

The optimal solution of the optimization problem that application Lagrange multiplier method obtains

��