A kind of flaw online test method being applicable to Woven textiles
Technical field
The present invention relates to a kind of flaw online test method being applicable to Woven textiles, belong to textile flaw on-line real-time test field.
Background technology
In Modern Textile Industry, the automatic detection replacing human eye to carry out fabric defects with the detection technique of advanced person can improve detection efficiency, reduces labour, reduce labour intensity and the further quality improving fabric.In China, Fabric Detection major part has still been come by artificial vision, and in testing process, human eye vision exists deviation, and fidelity factor is lower and testing result unstable, incompatible with large-scale industrial production.In textile production, if can not Timeliness coverage flaw, some substandard products cloth will be produced, cause the waste of material and the energy.If installing fabric flaw vision detection system, find that flaw can process, to reduce the waste of material.
Discovery is retrieved through applicant, Chinese invention patent application CN102967606A, propose a kind of textile machine fabric defect visual inspection system, its principle of work is, utilize image recognition technology to carry out the feature extraction of cloth photo, after comparison, result is sent to host processing systems, complete the control to whole machine vision inspection distribution system through host processing systems, and in real time corresponding final process is made to the result of its feedback.This technology can detect fabric defects point in real time, and the number of giving the alarm, stop weaving loom running, guide workman to process current flaw, also effectively reduce labor strength and labor cost.
Detection image recognition technology being applied to fabric has become the customary means of this area, but existing recognition methods mostly is the direct conversion of universal method, and its specific aim is not strong, still there is undetected and false retrieval, can not meet the requirement of high-quality.The present invention identifies according to the flaw of texture feature to fabric of fabric emphatically.
Summary of the invention
Object of the present invention: the defect overcoming above-mentioned prior art, proposes a kind of flaw online test method being applicable to Woven textiles newly, is carry out flaw identification for the feature of fabric, makes undetected, fallout ratio reduction.
The flaw online test method being applicable to Woven textiles that the present invention proposes, it is characterized in that: above cloth working direction, interval is provided with upstream camera, downstream camera, the shooting time arranging two cameras according to cloth pace and upstream and downstream camera pitch is poor, guarantee that two cameras can photograph the same region of Woven textiles, obtain continuous print some to image, this method includes following steps:
1st step: utilize upstream, downstream camera captured in real-time Woven textiles, and by paired image transfer to computing machine;
2nd step: computing machine carries out pre-service to the image received, and comprising: be normalized gray level image, and strengthen picture contrast by gray scale stretching;
3rd step: to the image zooming-out textural characteristics vector of upstream camera shooting, concrete steps are as follows:
A1, dual-tree complex wavelet transform is carried out to image, obtain the matrix that 6 256 are taken advantage of 256;
A2,6 the matrixes substitution equine husband models that will obtain, ask for the parameter of equine husband model, and be normalized the parameter obtained;
A3, to after normalization parameter build vector, obtain the textural characteristics vector of fabric in this image;
4th step: the textural characteristics vector obtain the 3rd step and the textural characteristics vector of specification product compare, if both Euclidean distances are not more than default specification product threshold value, then the fabric in this image is specification product, goes to the 1st step; Otherwise by the image centrally Linear cut that downstream camera is taken, obtain two pieces of subimages;
5th step: utilize the method for the 3rd step to extract the textural characteristics vector of two pieces of subimages respectively, and respectively the textural characteristics vector of two pieces of subimages and the textural characteristics vector of specification product are compared, if both Euclidean distances are all not more than default specification product threshold value, fabric then in this image is specification product, then go to the 1st step; If both Euclidean distances are greater than specification product threshold value, then corresponding subimage textural characteristics vector is compared with the woven fabric textural characteristics vector of all preliminary election flaw kinds respectively, if both Euclidean distances are not more than corresponding flaw threshold value, woven fabric then in corresponding subimage belongs to corresponding flaw kind, and computing machine carries out record; If both Euclidean distances are all greater than corresponding flaw threshold value, then the woven fabric in corresponding subimage is other flaw kinds, and computing machine carries out record, sends stop instruction according to above-mentioned flaw kind, and points out corresponding personnel immediately to process.
The present invention further improves as follows:
1, the above-mentioned flaw online test method being applicable to Woven textiles, the preliminary election flaw kind in described 5th step comprises: lack warp, crapand, stain, heavy warp, heavy latitude, broken hole.
2, the textural characteristics vector of described specification product and specification product threshold value determination method as follows:
B1, the Woven textiles sample choosing at least 200 specification product carry out image acquisition;
B2, utilize the method in described 3rd step to extract the textural characteristics vector of each image, the cluster centre of all specification product image texture characteristic vectors is specification product textural characteristics vector center vector;
The textural characteristics vector of b3, all specification product images is specification product threshold value to the maximal value of the Euclidean distance of described specification product textural characteristics vector center vector.
3, the woven fabric textural characteristics vector of flaw and corresponding flaw Threshold as follows:
C1, choose at least 200 and specify the Woven textiles sample of flaw category to carry out image acquisition;
C2, utilize the method in described 3rd step to extract the textural characteristics vector of each image, the cluster centre of all appointment flaw category image texture characteristic vectors is the center vector of corresponding flaw product textural characteristics vector;
The textural characteristics vector of c3, such flaw product images all is corresponding flaw threshold value to the maximal value of the Euclidean distance of corresponding flaw product textural characteristics vector.
4, the shooting time difference T=S/V of two cameras, in formula, S is the distance between camera shooting center, and V is the pace of Woven textiles, and downstream camera shooting time is later than upstream camera shooting time.
The present invention is directed to the texture feature of Woven textiles, the feature extracting method of dual-tree complex wavelet and equine husband models coupling is utilized to carry out texture blending, compare traditional texture blending method, can fast, effectively, accurately extract the textural characteristics of Woven textiles, and the textural characteristics extracted more is conducive to identifying, the results showed the accuracy significantly improving identification.The present invention utilizes dual camera to carry out fabric collection, first to upstream camera take the photograph region monoblock cloth carry out texture blending, whether Preliminary detection is specification product, again downstream camera is taken the photograph cloth subregion secondary texture blending, and carry out flaw identification if defective.The present invention uses two cameras to take same cloth at diverse location dexterously; avoid disturbing due to reflective etc. the erroneous judgement caused; subregional secondary identification is carried out after preliminary identification; substantially increase identification accuracy; detection speed can be ensured again simultaneously; take resource less, on-line checkingi can be adapted to, do not shut down detection.
To sum up, the present invention greatly reduces identification error rate, improves the accuracy of flaw database, contributes to the producer and adjust to improve product quality to weaving parameter.
Accompanying drawing explanation
Below in conjunction with accompanying drawing, the present invention is further illustrated.
Fig. 1 is the flow diagram of the inventive method.
Embodiment
Below in conjunction with the drawings and specific embodiments, the present invention will be further described.
The present embodiment is applicable to the flaw online test method of Woven textiles, its improvement is: above cloth working direction, interval is provided with upstream camera, downstream camera, according to cloth pace and on, downstream camera pitch arranges the shooting time difference T of two cameras, guarantee that two cameras can photograph the same region of Woven textiles, shooting time difference T=S/V, in formula, S is the distance between camera shooting center, V is the pace of Woven textiles, downstream camera shooting time is later than upstream camera shooting time, obtain continuous print some to image, the present embodiment method flow diagram is shown in Fig. 1, and concrete steps are as follows:
1st step: utilize upstream, downstream camera captured in real-time Woven textiles, and by paired image transfer to computing machine;
2nd step: computing machine carries out pre-service to the image received, and comprising: be normalized gray level image, and strengthen picture contrast by gray scale stretching;
3rd step: to the image zooming-out textural characteristics vector of upstream camera shooting, concrete steps are as follows:
A1, dual-tree complex wavelet transform is carried out to image, obtain the matrix that 6 256 are taken advantage of 256;
A2,6 the matrixes substitution equine husband models that will obtain, ask for the parameter of equine husband model, and be normalized the parameter obtained;
A3, to after normalization parameter build vector, obtain the textural characteristics vector of fabric in this image;
4th step: the textural characteristics vector obtain the 3rd step and the textural characteristics vector of specification product compare, if both Euclidean distances are not more than default specification product threshold value, then the fabric in this image is specification product, goes to the 1st step; Otherwise by the image centrally Linear cut that downstream camera is taken, obtain two pieces of subimages;
5th step: utilize the method for the 3rd step to extract the textural characteristics vector of two pieces of subimages respectively, and respectively the textural characteristics vector of two pieces of subimages and the textural characteristics vector of specification product are compared.If both Euclidean distances are all not more than default specification product threshold value, then the fabric in this image is specification product, then go to the 1st step.If both Euclidean distances are greater than specification product threshold value, then corresponding subimage textural characteristics vector is compared with the woven fabric textural characteristics vector of all preliminary election flaw kinds respectively, if both Euclidean distances are not more than corresponding flaw threshold value, woven fabric then in corresponding subimage belongs to corresponding flaw kind, and computing machine carries out record; If both Euclidean distances are all greater than corresponding flaw threshold value, then the woven fabric in corresponding subimage is other flaw kinds, and computing machine records flaw kind and sends stop instruction, and points out corresponding personnel immediately to process.
Preliminary election flaw kind in described 5th step comprises: lack warp, crapand, stain, heavy warp, heavy latitude, broken hole.
Textural characteristics vector and the specification product threshold value determination method of this enforcement specification product are as follows:
B1, the Woven textiles sample choosing at least 200 specification product carry out image acquisition;
B2, utilize the method in described 3rd step to extract the textural characteristics vector of each image, the cluster centre of all specification product image texture characteristic vectors is specification product textural characteristics vector center vector;
The textural characteristics vector of b3, all specification product images is specification product threshold value to the maximal value of the Euclidean distance of described specification product textural characteristics vector center vector.
Woven fabric textural characteristics vector and the corresponding flaw Threshold of the present embodiment flaw are as follows:
C1, choose at least 200 and specify the Woven textiles sample of flaw category to carry out image acquisition;
C2, utilize the method in described 3rd step to extract the textural characteristics vector of each image, the cluster centre of all appointment flaw category image texture characteristic vectors is the center vector of corresponding flaw product textural characteristics vector;
The textural characteristics vector of c3, such flaw product images all is corresponding flaw threshold value to the maximal value of the Euclidean distance of corresponding flaw product textural characteristics vector.
In addition to the implementation, the present invention can also have other embodiments.All employings are equal to the technical scheme of replacement or equivalent transformation formation, all drop on the protection domain of application claims.