CN101063660B - Method for detecting textile defect and device thereof - Google Patents

Abstract

This invention discloses one fabric deficiency test method and its device, which comprises the following steps: acquiring fabric production line surface images through CCD camera; using wavelet to change image for one layer of analysis into four sub ones non-overlapped as for LH, HL, HH and LL separately responding to the self-adaptive value process to realize the fabric deficiency positioning; then overlapping the three pairs of sub images to get the final position. This invention device comprises CCD cameral array, computer servo and its inside fabric deficiency test program and connection CCD camera array and computer servo and the CCD camera control line.

Description

A kind of method for detecting textile defect and device thereof

Technical field

The present invention relates to a kind of method for detecting textile defect and device thereof, belong to technical field of quality detection.

Background technology

Quality testing is the important step that modern industry is produced, and occupies critical role in process of production.The commercial production that with the textile is representative is all the more so.Since the entry to WTO, the textile industry of China is facing the macro environment of wider and higher level opening, will obtain more opportunity to develop, but also can welcome more arduous challenge.China is as traditional textile production and big export country, and textile industry is seized of extremely important status in national economy, and in considerable time, remains one of important export strong industry.Keep textile industry to continue, stable development is all most important to the development of China's foreign trade cause and even whole national economy.Yet,, human more and more higher to the quality requirements of textile along with the raising day by day of living standards of the people.Therefore, strengthen detection to textile, strengthen the synthesized competitiveness of China's textile, promote that textile industry continues, stable development is of great practical significance improving the total quality of China's textile.

At present traditional method for detecting textile defect is by manually carrying out, and the eyes that promptly depend on the people remove to find product defects.Human eye can be found the defective locations of textile exactly, but because the dullness of testing itself, dull, and people's eyes are easy to produce fatigue, and along with the increase of detection time, people's eyes are tired more, and the error of generation can increase sharply.According to statistics, even skilled detection person, dependence is manually carried out the efficient of textile defect detection and also is difficult to reach 70%.In addition, limit by the physiological function of human eye, testing be slowly, time-consuming, be difficult to satisfy the quick requirement of modernized caused by spinning industrial production.

Summary of the invention

The object of the present invention is to provide a kind of based on Flame Image Process, quick, real-time textile defect automatic testing method, thereby substitute traditional manual detection method, with deficiency and the defective that remedies prior art.

Another object of the present invention is to provide a kind of textile defect automatic detection device based on said method.

The inventive method realizes by following steps: at first obtain fabric face image on the textile production line by ccd video camera, use wavelet transformation image is carried out one deck wavelet decomposition, with picture breakdown is four subgraphs of non-overlapping copies, promptly, LH subgraph (high-frequency information of expression horizontal direction), HL subgraph (high-frequency information of expression vertical direction), HH subgraph (high-frequency information of expression diagonal), LL subgraph (the approximate component of presentation video).Utilize computer server and inner textile defect trace routine module thereof, respectively to LH, HL, LL subgraph carry out adaptive threshold and handle and then realize the fabric defect location, and folded then triple raise subgraph obtains final defective locations.Therefore, the subgraph that obtains after the stack comprises the horizontal edge that the position appears in fabric defect, vertical edge, and the global shape information of defective.

The inventive system comprises ccd video camera array, array of source, computer server and connect ccd video camera array and computer server data line, be connected ccd video camera control line and light source control line on the computer server.In addition, computer server also comprises the textile defect trace routine module that it is inner.

The present invention adopts the wavelet analysis technology that has been widely used in various fields, the analytical approach that wavelet analysis provides a kind of adaptive time domain and frequency domain to localize simultaneously, no matter analysing low frequency or high frequency partial signal, when it can both be regulated automatically-the frequency window, to adapt to the needs of actual analysis.Wavelet analysis has very strong dirigibility in local time-frequency analysis, can focus on any details of signal time slot, when being called-and the microscope of frequency analysis.Therefore, at the texture of the texture of fabric face defect area and normal fabric face on every side intensity profile different in image, after textile image carried out wavelet decomposition, utilize the suitable feature extracting method, can realize fully the defective locations of textile is accurately located, thereby textile defect is carried out fast detecting.That is, can not only judge that fabric face has zero defect, the particular location and the shape information of fault can also be provided.

Description of drawings

Fig. 1 is an one-piece construction synoptic diagram of the present invention.

Fig. 2 is a fabric defect testing process process flow diagram of the present invention.

The fabric face defective schematic images that Fig. 3 obtains for ccd video camera.

Fig. 4 is the effect synoptic diagram after the present invention handles HL subgraph threshold process.

Fig. 5 is the effect synoptic diagram after the present invention handles LH subgraph threshold process.

Fig. 6 is the effect synoptic diagram after the present invention handles LL subgraph threshold process.

Fig. 7 is the effect synoptic diagram of the present invention after to three secondary subgraph overlap-add procedure.

Wherein, 1. textile production line, 2. array of source, 3.CCD video camera array, 4. data line, 5. computer server, 6.CCD video camera control line, 7. light source control line.

Embodiment

Further specify the present invention below in conjunction with accompanying drawing and by specific embodiment.

As shown in Figure 1, the inventive system comprises ccd video camera array 3, array of source 2, computer server 5 and inner textile defect trace routine module thereof, connect ccd video camera array 3 and computer server 5 data line 4, be connected ccd video camera control line 6 and light source control line 7 on the computer server 5.The present invention to the fabric defect detection method is: at first obtain fabric face image on the textile production line by ccd video camera 3, and it is carried out one deck wavelet decomposition, with four subgraphs of picture breakdown, be expressed as respectively: LH subgraph (detail of the high frequency of expression level), HL subgraph (representing vertical detail of the high frequency), HH subgraph (detail of the high frequency on the expression diagonal line), LL subgraph (the approximate component of presentation video).Utilize computer server and inner textile defect trace routine module thereof, respectively to LH, HL, LL subgraph carry out adaptive threshold and handle and then realize the fabric defect location, and folded then triple raise subgraph obtains final defective locations.

Textile defect trace routine detailed process is that example describes with HL subgraph defect inspection process.Suppose c (j, the k) value of expression HL subgraph every bit correspondence, i.e. the wavelet coefficient of the HL subgraph every bit correspondence that obtains later on of wavelet decomposition, wherein, j, k are the coordinate figure of image, M, N are respectively HL subgraph columns and line number.The specific implementation step is as follows:

1. use template $\left(\begin{array}{ccc}-1& 0& 1\\ -1& 0& 1\\ -1& 0& 1\end{array}\right)$ The HL subgraph is carried out convolution, to strengthen the contrast of vertical high frequency detailed information, with h (j, k) the pairing convolution results of expression HL subgraph every bit.(indicate: not optional can choosing of this step, at the tangible textile product of part texture, this step can omit).

2. at each row of HL subgraph, each coefficient is carried out normalization:

$P_{\mathrm{jk}}=\frac{\left|h(j,k)\right|}{\underset{k=1}{\overset{N}{\mathrm{\Σ}}}\left|h(j,k)\right|};$

Calculating this journey level changes:

${\mathrm{Dir}}_j=\frac{1}{N}\underset{k=1}{\overset{\mathrm{<figure-callout\; id="2"\; label="N\; p\; jk"\; filenames="S07113452320070306E000011.png"\; state="\{\{state\}\}">N</figure-callout>}}{\mathrm{\&Sigma;}}}{p_{\mathrm{jk}}}^{}{{}_{}}^2;$

3. each row from left to right calculates d the adjacent capable level of putting of each some left and right sides and is changed to:

${\mathrm{Dir}}_{j\&prime;k}=\frac{1}{2d+1}\underset{i=k-d}{\overset{k+\mathrm{<figure-callout\; id="2"\; label="d\; p\; ji"\; filenames="S07113452320070306E000011.png"\; state="\{\{state\}\}">d</figure-callout>}}{\mathrm{\&Sigma;}}}{p_{\mathrm{ji}}}^{}{{}_{}}^2;$

Wherein, d can get the value between the 2-4.

4. carry out threshold process at each row:

Wherein, σ _jBe j row coefficient p _JkMean square deviation.δ is a sensitivity coefficient, can get δ=3.The image of Fig. 4 for obtaining after the threshold value, image has reflected the horizontal edge detailed information of fabric defect.

Same, use template $\left(\begin{array}{ccc}1& 1& 1\\ 0& 0& 0\\ -1& -1& -1\end{array}\right)$ The LH subgraph is carried out convolution, with the contrast of enhanced level detail of the high frequency.(indicate: not optional can choosing of this step, at the tangible textile product of part texture, this step can omit).At each row of LH subgraph, to the processing of HL subgraph, different was that the processing of LH subgraph is undertaken by each row above processing procedure was similar.Obtain the image of threshold value after the processing, as shown in Figure 5.Image has reflected the vertical edge detailed information of fabric defect.

For the LL subgraph, need handle simultaneously its row and column.Use template $\left(\begin{array}{ccc}-1& 0& 1\\ -1& 0& 1\\ -1& 0& 1\end{array}\right)$ After the HL subgraph carried out convolution, duplicate in processing to the HL subgraph at processing procedure of each row.Same, use template $\left(\begin{array}{ccc}1& 1& 1\\ 0& 0& 0\\ -1& -1& -1\end{array}\right)$ After the LH subgraph carried out convolution, be similar to the processing that is directed to the LH subgraph at the processing procedure of each row, Fig. 6 is the result after the threshold value.

Because HH subgraph (high-frequency information on the diagonal line) comprises most of noise of image, therefore it is not handled, detect the accuracy of effect and the real-time of processing with raising.

At last, three secondary subgraphs are superposeed, obtain final defective locations, and the global shape information of defective, Fig. 7 is final result.At different application, take different subsequent treatment work, as reporting to the police mark defective locations etc.

Method of the present invention adopts wavelet analysis technology, only carries out one deck wavelet decomposition, and computing velocity is fast, and the real-time of handling has been considered in accurate positioning again.In addition, the present invention does not need to carry out machine learning or training without any need for the priori of textile defect, and versatility is good.

Device of the present invention is the optional equipment that is independent of outside the textile production line, the processing aspect that not only can be widely used in textile defect, and can apply to any defects detection with grain surface feature object, detect fields such as control such as surface quality of products such as timber, plastics, potteries.

Claims (1)

1. a method for detecting textile defect is characterized in that, may further comprise the steps: at first obtain fabric face image on the textile production line by ccd video camera; Use wavelet basis function image is carried out one deck wavelet decomposition, with picture breakdown is four subgraphs of non-overlapping copies, that is, LH subgraph, the high-frequency information of its expression horizontal direction, the HL subgraph, the high-frequency information of its expression vertical direction, HH subgraph, the high-frequency information of its expression diagonal, the LL subgraph, the approximate component of its presentation video; To LH, HL, LL subgraph carry out adaptive threshold and handle and then realize the fabric defect location respectively; Folded then triple raise subgraph obtains final defective locations; Describedly the HL subgraph is carried out adaptive threshold handle and may further comprise the steps:

(1) uses template

The HL subgraph is carried out convolution, to strengthen the contrast of vertical high frequency detailed information, with h (j, k) the pairing convolution results of expression HL subgraph every bit;

(2) at each row of HL subgraph, each coefficient is carried out normalization:

$p_{\mathrm{jk}}=\frac{\left|h(j,k)\right|}{\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}\left|h(j,k)\right|};$

Calculating this journey level changes:

${\mathrm{Dir}}_j=\frac{1}{N}\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}{p_{\mathrm{jk}}}^2;$

(3) each row from left to right calculates d the adjacent capable level of putting of each some left and right sides and is changed to:

${\mathrm{Dir}}_{j^{\&prime;}k}=\frac{1}{2d+1}\underset{i=k-d}{\overset{k+d}{\mathrm{\&Sigma;}}}{p_{\mathrm{ji}}}^2;$

Wherein, d gets the value between the 2-4;

(4) carry out threshold process at each row:

Wherein, σ _jBe j row coefficient p _JkMean square deviation, δ is a sensitivity coefficient, gets δ=3.

Images