CN104751443A - Cotton fault detecting and identifying method based on multi-spectrum technology - Google Patents

Abstract

The invention discloses a cotton fault detecting and identifying method based on a multi-spectrum technology. The method comprises the steps: (1) image acquiring; (2) image pre-processing; (3) the preliminary detection of faults; carrying out edge detection on an improved morphological gradient algorithm to obtain the comprehensive edge detection intensities of all faults to obtain an image with contrast enhancement of all faults; (4) precise location of faults: converting the image into a connected domain image through an improved iterative threshold and morphological processing, precisely locating the faults by selecting a main fault algorithm, and extracting fault areas by marking; (5) cotton ginning quality judgment: carrying out statistics on the amount of all faults, and calculating the total cotton faults to obtain a cotton ginning quality grade. The cotton fault detecting and identifying method provided by the invention has the advantages of being simple in theory, fast in detection speed and high in detection precision.

Description

Based on multispectral technology cotton defect detection and recognition methods

Technical field

The invention belongs to field of optical measuring technologies, particularly a kind of quality of roll for cotton automatic classification detection and Identification method.

Background technology

Gined cotton fault refers to the impurity of band fiber and hinders the large class of the fiber two of weaving, comprises brokenly seed, mote, rope silk, soft seed epidermis, stiff sheet, bearded mote and 7 kinds, cotton knot.Because fault is difficult to discharge in fabrication processes, residual impurity fault is wrapped in sliver or is attached to yam surface, and its consequence causes the dry deterioration of bar, and formation yarn nep, impurity increase, and yarn fine hair plumage increases.Its surface exhibits of embryo cloth caused with the yarn that bearded mote quantity is large goes out a large amount of cotton knots, impurity, and after dyeing, cloth cover will manifest infertile color dot, card nailing neps, and feel is stiff coarse, thus affects the development of textile industry, brings serious loss to national economy.Therefore, the quick detection of gined cotton fault content judges most important for raw cotton quality grade.

Quality of roll for cotton according to fault content respectively as well, in, difference third gear.Current most domestic enterprise generally adopts sight detection method to carry out the identification of fault.The test accuracy of sight detection method is often by the impact of human factor, and for tiny fault, false drop rate and the loss of testing result are higher, and time-consuming.Can produce a large amount of cotton-wools and airborne dust in testing process, such testing staff often can be subject to the impact of external environment simultaneously, and human eye is also easily tired through working long hours, and causes accuracy to decline.Along with the development of computer technology and photoelectric detecting technology, there is various cotton quality automatic testing method, but it is different from Target detection and identification mode generally, Full-automatic cotton defect detection and recognition system have high speed, high-precision requirement, and such cotton defect detection and recognition methods are faced with following main difficult technical:

(1) current art and method generally adopt visible ray as light source, can only effective identification division fault, and quality discrimination precision is on the low side.

(2) image disruption factor is many.Due to exist in imaging process under the wave characteristic of light source and industrial condition shade, the various interference such as light reflection and absorption, even its imaging results of lily cotton also may occur that light and shade is uneven, general detection method is made easily to be disturbed the impact of factor.

(3) cotton fault has seven types, complex shape, because this increasing detection difficulty.Cotton knot is similar to cotton morphological feature with Suo Si, and general simple target detection method therefore can not be used to identify.

Summary of the invention

The technical problem to be solved in the present invention is just: the technical matters existed for prior art, the invention provides a kind of principle is simple, automaticity is high, detection speed is fast, accuracy of detection is high gined cotton defect detection and recognition methods.

For solving the problems of the technologies described above, the present invention by the following technical solutions:

A kind of based on multispectral technology cotton defect detection and recognition methods, the steps include:

(1) Image Acquisition: the relation utilizing the different fault optimized image sharpness of cotton and illumination and driving voltage, fault optimized image is obtained by measuring illumination feedback driving light source, utilize White LED light source and multispectral camera first to obtain the coloured image of cotton fault, then utilize near infrared semiconductor line laser instrument and multispectral camera obtain cotton fault line image and line image are spliced into a width panoramic picture;

(2) Image semantic classification: comprise and carry out noise removal to image, extracts color camera image R, G, B tri-channel informations, obtains three and pass through gray level image;

(3) fault Preliminary detection: utilize improvement Morphological Gradient algorithm to carry out rim detection, obtain each fault overall edge detected intensity, obtain each fault contrast strengthen image;

(4) fault is accurately located: obtain defect regions binary image by improving iteration method, remove impurity point, to image expansion, corrosion and filling algorithm operation, image is converted to connected domain image, by selecting major defect algorithm accurately to locate fault, and break seed, mote and stiff sheet defect regions by marker extraction.For near-infrared image, deduct color camera process image and obtain defect regions, maximum variance between clusters is utilized to carry out Iamge Segmentation, Morphological Gradient algorithm carries out the detection of fault edge strength, to image expansion, corrosion and filling algorithm operation, image is converted to connected domain image, by selecting major defect algorithm accurately to locate fault, and by marker extraction rope silk and cotton knot defect regions;

(5) ginning quality judging: add up each fault quantity, calculates the total grain number of cotton fault, obtains ginning quality scale.

As a further improvement on the present invention:

The concrete steps of described step (2) are:

(2.1) the image F (i adopting medium filtering to gather step 1, j) with G (i, j) carry out noise remove and obtain image f (i, j) with g (i, j), the two dimension pattern plate in 5 × 5 regions is adopted, in template, pixel is greater than default threshold value 4500, then all pixel grey scales are set to 255, otherwise are 0, thus realize the object selecting major defect.Wherein 4500 choose be through test of many times after determined empirical value.In image, (i, j) some place exports med [x (i)] through medium filtering:

$\mathrm{med}\left[x\left(i\right)\right]=\left\{\begin{array}{c}x(k+1),i=2k+1\\ 0.5x\left(k\right)+0.5x(k+1),i=2k\end{array}\right.$

(2.2) information extraction of R, G, B triple channel is carried out to cotton fault coloured image F (i, j), obtain R passage gray level image f _r(i, j), G passage gray level image f _g(i, j), and channel B gray level image f _b(i, j).

The concrete steps of described step (3) are:

(3.1) respectively rim detection is carried out for coloured image R, G, B triple channel gray-scale map, R channel edge detected intensity grad (f _r) be:

$\mathrm{grad}\left(f_R\right)=f_R\⊗g-f_R\⊕g;$

G channel edge detected intensity grad (f _g) be:

$\mathrm{grad}\left(f_G\right)=f_G\⊗g-f_G\⊕g$

Channel B rim detection strength g rad (f _b) be:

$\mathrm{grad}\left(f_B\right)=f_B\⊗g-f_B\⊕g$

Wherein f _r, f _gand f _bfor R, G, B component, g is structural element.Adopt 3 × 3 diamond structures as structural element, if rim detection intensity is greater than given threshold value, then this pixel is the edge pixel point obtained.

(3.2) the broken seed in cotton fault and the edge strength operator of bearded mote are:

Grad (f _{broken seed})=grad (f _r)+grda (f _g)-grad (f _b)

According to many experiments default threshold value T _seed=315, what be greater than threshold value is broken seed, and what be less than threshold value is bearded mote.

In cotton fault, the edge strength operator of dead cotton is:

grad(f)＝grad(f _R)+grad(f _G)+grad(f _B)×2.5

Grad (f _{dead cotton})=grad (f)-grad (f _{broken seed})

The concrete steps of described step (4) are:

(4.1) utilize cotton to break seed, mote and dead cotton overall edge detected intensity and obtain maximum gradation value t_max and minimum gradation value t_min, initial threshold is $T^0=\frac{t\_\max +t\_\min }{2};$

(4.2) according to threshold value T ^kpseudo-gray level image is divided into target and background two parts, and target average gray value is::

$Z_0=\frac{\underset{z(i,j)<T^k}{\mathrm{\&Sigma;}}z(i,j)n(i,j)}{\underset{z(i,j)<T^k}{\mathrm{\&Sigma;}}n(i,j)}$

Background average gray value is:

$Z_B=\frac{\underset{z(i,j)<T^k}{\mathrm{\&Sigma;}}z(i,j)n(i,j)}{\underset{z(i,j)<T^k}{\mathrm{\&Sigma;}}n(i,j)}$

The wherein gray scale put for (i, j) on shade of gray image of z (i, j); N (i, j) is the weight coefficient that (i, j) puts.

(4.3) obtain according to background average gray value and target average gray value the new threshold value being applicable to raw cotton defect detection:

T ^k+1＝n(Z _O+Z _B)

Optimal threshold is obtained according to new threshold value:

T _op＝T _k；|T _k+1-T _k|≤0.01

Wherein, k=1,2,3..., 8, T _kfor n _kthreshold value during=0.1 × k, according to test of many times result for broken seed and mote n=0.6, for dead cotton n=0.5.

(4.4) according to maximum variance between clusters, near-infrared image is divided into background area and target area, utilizes edge detection operator to obtain cotton knot and Suo Si fault overall edge intensity:

$\mathrm{grad}\left(f\right)=f\&CircleTimes;g-f\&CirclePlus;g$

(4.5) major defect is selected:

(4.5.1) according to cotton defect image feature, setting area threshold 80 removes image f ₁assorted point in (i, j), 80 be chosen for determined empirical value after test of many times.

(4.5.2) assorted removal image expanded, corrode and filling algorithm operation, obtain process image f ₃(i, j):

$f\&CircleTimes;g(x,y)=\underset{(i,j)}{\max }\{f(x-i,y-j)+g(i,j)\}$

$f\&CircleTimes;g(x,y)=\underset{(x,y)}{\max }\{f(x-i,y+j)+g(i,j)\}$

(4.5.3) image g is converted to connected domain image g ₁(i, j): first Horizon Search is done to image g, if horizontal adjacent 26th of (i, j) point is white point, then 2 are middle is a little white point, otherwise does not do operation continuation search; Then image g does longitudinal searching, if longitudinal adjacent 9th of (i, j) point is white point, then 2 middle is a little white point, otherwise does not do operation continuation search.26 and 9 experimentally in determined empirical value after width and cotton fault feature test of many times in structure light image.

(4.6) cotton fault is extracted:

(4.6.1) to image f ₄in region carry out marking each region and use an integer from 1 to mark, different defect regions mark value is different, background area uses 0 to mark, marked region is greater than 7000 and obtains as dead cotton, be greater than 150 to be less than 7000 simultaneously and to be judged to be brokenly seed, all the other are judged to be mote, 7000 and 150 be chosen for test of many times after determined empirical value.

(4.6.2) to image g ₁in region carry out marking each region and use an integer from 1 to mark, different defect regions mark value is different, and background area uses 0 to mark, and marked region is greater than 600 for rope silk, be less than 600 for cotton knot, 600 be chosen for test of many times after determined empirical value.

Compared with prior art, the invention has the advantages that:

(1) detection and Identification speed of the present invention is fast, ensure that the rate request of cotton defect detection and cotton ginning quality Identification.The present invention is different from general detection method, the imaging of different fault difference is adopted to extract the algorithm of defect regions, thus reach the non-defect regions of removal to the impact of testing result and different defect regions extraction interference problem, the saving algrithm time, substantially increase detection speed.

(2) accuracy of detection of the present invention is high.Because cotton fault is of a great variety, complex shape, dissimilar fault gray feature is very different, and one often there is multiple fault in width cotton defect image simultaneously, be different from general target detection box identification single, of the same type, therefore how effectively faults all on image to be detected, be a difficult point simultaneously.The present invention utilizes multispectral camera can obtain the characteristic of coloured image and near-infrared image simultaneously, improvement Morphological Gradient algorithm and improvement iteration method are combined by coloured image, detect brokenly seed, mote and stiff sheet fault well, for near-infrared image, maximum variance between clusters and Morphological Gradient algorithm is utilized to identify Suo Si and cotton knot fault well, thus identify each fault of cotton well simultaneously, solve this problem.

(3) the present invention determines shelves requirement according to the cotton quality of technician's testing result and reality, devise each fault weight shared by quality evaluation, obtain cotton fault sum computing formula, obtain ginning quality scale according to the total grain number of fault, further ensure the accuracy of testing result.

Accompanying drawing explanation

Fig. 1 provided by the invention one to grow cotton defect detection method overall procedure schematic diagram.

Fig. 2 a is the fault coloured image taken in embody rule example.

Fig. 2 b takes fault single width near infrared ray structure light image in embody rule example.

Fig. 2 c is the panoramic picture of near infrared ray structure light image splicing in embody rule example.

Fig. 3 a is that cotton fault coloured image G passage extracts image in embody rule example.

Fig. 3 b is that cotton fault coloured image channel B extracts image in embody rule example.

Fig. 3 c is that cotton fault coloured image R passage extracts image in embody rule example.

Fig. 4 a is that innovatory algorithm extracts broken seed image in embody rule example.

Fig. 4 b is that innovatory algorithm extracts mote image in embody rule example.

Fig. 4 c is that innovatory algorithm extracts stiff picture in embody rule example.

Fig. 5 a is the rear image of segmentation in embody rule example.

Fig. 5 b is that assorted point removes rear image in embody rule example.

Fig. 5 c is image after Morphological scale-space in embody rule example.

Fig. 6 a is the rear image of near infrared cotton defect image segmentation in embody rule example.

Fig. 6 b is image after being communicated with after the segmentation of near infrared cotton defect image in embody rule example.

Fig. 7 a is cotton knot fault final detection result in embody rule example.

Fig. 7 b is rope silk fault final detection result in embody rule example.

Embodiment

For making object of the present invention, technical scheme and technique effect clearly; below in conjunction with the specific embodiment of the invention and respective drawings; technical scheme in the embodiment of the present invention is clearly and completely described; but following examples can not be interpreted as to of the present invention can the restriction of practical range; based on the embodiment in the present invention; other embodiments all that those of ordinary skill in the art obtain under the prerequisite not making creative work, all belong to the scope of protection of the invention.

As shown in Figure 1, of the present invention based on multispectral technology cotton defect detection and recognition methods, its flow process is:

1. Image Acquisition: obtain cotton sample to be measured, detection system obtains cotton multispectral image, obtain fault coloured image F (i, j), see Fig. 2 (a) and fault single width near infrared ray structure light image, see Fig. 2 (b), move cotton samples by guide rail and obtain N frame near infrared ray image, N frame near-infrared image is spliced into width panoramic picture G (i, j), see Fig. 2 (c).

2. Image semantic classification: due to the impact of the factors such as Test Field, can introduce noise to the Cotton Images collected, this noise is mainly salt-pepper noise.

2.1 present invention employs the image F (i that medium filtering gathers step 1, j) with G (i, j) carry out noise remove and obtain image f (i, j) with g (i, j), the two dimension pattern plate in 5 × 5 regions is adopted, in template, pixel is greater than default threshold value 4500, then all pixel grey scales are set to 255, otherwise are 0, thus realize the object selecting major defect.Wherein 4500 choose be through test of many times after determined empirical value.

2.2 extract color camera image R, G, B tri-channel informations, obtain R passage gray level image f _r(i, j), see Fig. 3 (a), G passage gray level image f _g(i, j), see Fig. 3 (b) and channel B gray level image f _b(i, j), see Fig. 3 (c).

3. couple image f _r(i, j), f _g(i, j) and f _b(i, j) improvement Morphological Gradient algorithm is utilized to carry out rim detection, obtain each fault overall edge detected intensity, obtain cotton and break seed fault enhancing image, see Fig. 4 (a), cotton mote fault strengthens image, strengthens image, see Fig. 4 (c) see Fig. 4 (b) and the stiff sheet fault of cotton; Segmentation image f is obtained by improving iteration method ₁(i, j), see Fig. 5 (a), setting area threshold 80 removes impurity point, and assorted point removes image f ₂(i, j) see Fig. 5 (b), to image f ₂(i, j) carries out expanding, corroding and filling algorithm operation, process image f ₃(i, j) is see Fig. 5 (c).Wherein 80 choose be through test of many times after determined empirical value.

4. the region in pair image marks, each region uses an integer from 1 to mark, different defect regions mark value is different, background area uses 0 to mark, marked region is greater than 7000 and obtains as dead cotton, be greater than 150 to be less than 7000 simultaneously and to be judged to be brokenly seed, all the other are judged to be mote.Wherein 150 and 7000 choose be through test of many times after determined empirical value.

5., for near-infrared image g (i, j), deduct color camera process image f (i, j) cotton rope silk and cotton knot defect regions is obtained, utilize maximum variance between clusters to carry out Iamge Segmentation, segmentation image g (i, j) is see Fig. 6 (a).Image g (i, j) is converted to connected domain image g ₁(i, j), see Fig. 6 (b).Mark region in image, each region uses an integer from 1 to mark, and different cotton defect regions mark value is different, and background area uses 0 to mark, and marked region is greater than 600 for rope silk, is less than 600 for cotton knot.As shown in Fig. 7 (a) He Fig. 7 (b), be through the image after Precision Orientation Algorithm process, wherein Fig. 7 (a) is cotton knot defect detection result, and Fig. 7 (b) is rope silk defect detection result.Wherein 600 choose be through test of many times after determined empirical value.

6. add up each fault quantity, wherein calculate the total grain number of cotton fault, obtain ginning quality scale.

Below be only the preferred embodiment of the present invention, protection scope of the present invention be not only confined to above-described embodiment, all technical schemes belonged under thinking of the present invention all belong to protection scope of the present invention.It should be pointed out that for those skilled in the art, some improvements and modifications without departing from the principles of the present invention, should be considered as protection scope of the present invention.

Claims (4)

1., based on multispectral technology cotton defect detection and a recognition methods, the steps include:

(1) Image Acquisition: the relation utilizing the different fault optimized image sharpness of cotton and illumination and driving voltage, fault optimized image is obtained by measuring illumination feedback driving light source, utilize White LED light source and multispectral camera first to obtain the coloured image of cotton fault, then utilize near infrared semiconductor line laser instrument and multispectral camera obtain cotton fault line image and line image are spliced into a width panoramic picture;

(2) Image semantic classification: comprise and carry out noise removal to image, extracts color camera image R, G, B tri-channel informations, obtains three and pass through gray level image;

(3) fault Preliminary detection: utilize improvement Morphological Gradient algorithm to carry out rim detection, obtain each fault overall edge detected intensity, obtain each fault contrast strengthen image;

(4) fault is accurately located: obtain defect regions binary image by improving iteration method, remove impurity point, to image expansion, corrosion and filling algorithm operation, image is converted to connected domain image, by selecting major defect algorithm accurately to locate fault, and break seed, mote and stiff sheet defect regions by marker extraction.For near-infrared image, deduct color camera process image and obtain defect regions, maximum variance between clusters is utilized to carry out Iamge Segmentation, Morphological Gradient algorithm carries out the detection of fault edge strength, to image expansion, corrosion and filling algorithm operation, image is converted to connected domain image, by selecting major defect algorithm accurately to locate fault, and by marker extraction rope silk and cotton knot defect regions;

(5) ginning quality judging: add up each fault quantity, calculates the total grain number of cotton fault, obtains ginning quality scale.As a further improvement on the present invention.

2. according to claim 1 based on multispectral technology cotton defect detection and recognition methods, it is characterized in that the concrete steps of step (2):

(2.1) the image F (i adopting medium filtering to gather step 1, j) with G (i, j) carry out noise remove and obtain image f (i, j) with g (i, j), the two dimension pattern plate in 5 × 5 regions is adopted, in template, pixel is greater than default threshold value 4500, then all pixel grey scales are set to 255, otherwise are 0, thus realize the object selecting major defect.Wherein 4500 choose be through test of many times after determined empirical value.In image, (i, j) some place exports med [x (i)] through medium filtering:

(2.2) information extraction of R, G, B triple channel is carried out to cotton fault coloured image F (i, j), obtain R passage gray level image f _r(i, j), G passage gray level image f _g(i, j), and channel B gray level image f _b(i, j).

3. according to claim 1 based on multispectral technology cotton defect detection and recognition methods, it is characterized in that the concrete steps of step (3):

(3.1) respectively rim detection is carried out for coloured image R, G, B triple channel gray-scale map, R channel edge detected intensity grad (f _r) be:

G channel edge detected intensity grad (f _g) be:

Channel B rim detection strength g rad (f _b) be:

Wherein f _r, f _gand f _bfor R, G, B component, g is structural element.Adopt 3 × 3 diamond structures as structural element, if rim detection intensity is greater than given threshold value, then this pixel is the edge pixel point obtained.

(3.2) the broken seed in cotton fault and the edge strength operator of bearded mote are:

Grad (f _{broken seed})=grad (f _r)+grad (f _g)-grad (f _b)

According to many experiments default threshold value T _seed=315, what be greater than threshold value is broken seed, and what be less than threshold value is bearded mote.

In cotton fault, the edge strength operator of dead cotton is:

grad(f)＝grad(f _R)+grad(f _G)+grad(f _B)×2.5

Grad (f _{dead cotton})=grad (f)-grad (f _{broken seed}).

4. according to claim 1 based on multispectral technology cotton defect detection and recognition methods, it is characterized in that the concrete steps of step (4):

(4.1) utilize cotton to break seed, mote and dead cotton overall edge detected intensity and obtain maximum gradation value t_max and minimum gradation value t_min, initial threshold is

(4.2) according to threshold value T ^kpseudo-gray level image is divided into target and background two parts, and target average gray value is::

Background average gray value is:

The wherein gray scale put for (i, j) on shade of gray image of z (i, j); N (i, j) is the weight coefficient that (i, j) puts.

(4.3) obtain according to background average gray value and target average gray value the new threshold value being applicable to raw cotton defect detection:

T ^k+1＝n(Z _O+Z _B)

Optimal threshold is obtained according to new threshold value:

T _op＝T _k；|T _k+1-T _k|≤0.01

Wherein, k=1,2,3..., 8, T _kfor n _kthreshold value during=0.1 × k, according to test of many times result for broken seed and mote n=0.6, for dead cotton n=0.5.

(4.4) according to maximum variance between clusters, near-infrared image is divided into background area and target area, utilizes edge detection operator to obtain cotton knot and Suo Si fault overall edge intensity:

(4.5) major defect is selected:

(4.5.1) according to cotton defect image feature, setting area threshold 80 removes image f ₁assorted point in (i, j), 80 be chosen for determined empirical value after test of many times.

(4.5.2) assorted removal image expanded, corrode and filling algorithm operation, obtain process image f ₃(i, j):

(4.5.3) image g is converted to connected domain image g ₁(i, j): first Horizon Search is done to image g, if horizontal adjacent 26th of (i, j) point is white point, then 2 are middle is a little white point, otherwise does not do operation continuation search; Then image g does longitudinal searching, if longitudinal adjacent 9th of (i, j) point is white point, then 2 middle is a little white point, otherwise does not do operation continuation search.26 and 9 experimentally in determined empirical value after width and cotton fault feature test of many times in structure light image.

(4.6) cotton fault is extracted:

(4.6.1) to image f ₄in region carry out marking each region and use an integer from 1 to mark, different defect regions mark value is different, background area uses 0 to mark, marked region is greater than 7000 and obtains as dead cotton, be greater than 150 to be less than 7000 simultaneously and to be judged to be brokenly seed, all the other are judged to be mote, 7000 and 150 be chosen for test of many times after determined empirical value.

(4.6.2) to image g ₁in region carry out marking each region and use an integer from 1 to mark, different defect regions mark value is different, and background area uses 0 to mark, and marked region is greater than 600 for rope silk, be less than 600 for cotton knot, 600 be chosen for test of many times after determined empirical value.