Research On Intelligent Defect Detection Methods Of Vehicle Leather

As living standards improve,having a private car becomes more and more common,people are not only concerned about the car's performance,but also set their sights on the car's interior quality.Leather as an important material of car's major interior components such as seats and other leather parts,its quality are strictly required.But the mosquito bites and human accidently hurt in the period growing and processing of leather raw materials,making it inevitably existence of various defects.So that quickly and accurately locating the defects on leather surface has a very important significance for improving product quality and guiding subsequent processing in manufacturing production.Since artificial detection currently used in car seat manufacturers has the miss rate,low efficiency and the feasibility and advantages of the method based on computer vision detection and make it hopes to introduce solutions of surface defects detection based on computer vision to replace the machine labor,which making the manufacturing chain safer,more efficiency,more stable,more objective and more costs.Facing the problems that the detection extremely difficult and slow to leather,due to the low contrast between the defects and non-defects regional and the interference of random texture in image,and there is no objective and quantitative evaluation method for a particular local effect of defect detection.Through the preliminary analysis of the leather defect samples,this article has done the following areas of research for leather flaws automatic detection which is difficult to find by artificial vision:First of all,the method of evaluation in flaw detection was established.Regarded the flaw detection as a special classification,an evaluation system for quantitative evaluation calculation has been proposed based on recall and accuracy,reference text classification evaluation index.First is to obtain the gold standard by a way of recognition and segmentation in the sample image which its defect area marked artificial as a reference to algorithms result.Then is to calculate comprehensive evaluation parameters defined,such as P,R,f,F.So it can digital evaluate the detection effect at the pixel-level,and provide objective guidance for the research of algorithms.Secondly,flaws detection model based on visual saliency degree is put forward,based the result of research that leather defect is a form of human eye attention and selection mechanism.Defect detection algorithm for leather image based on the model,comparing features of color and brightness to image itself as a template is used to calculate a saliency figure.And according to characteristic the saliency higher of the "highlights" prat in uniform random distribution texture image,defect area can detect using the segmentation method of region growing by the most saliency point.Experiments showed that the algorithm solves the problem of difficulty in template or effective feature extraction and could effectively detect tiny leather defects,by comparing this algorithm and existing defects detection algorithm based on threshold segmentation,FCM or SVM applied to leather defects detection.Finally,we devised a texture expression based on gray level co-occurrence matrix based on the research of texture analysis technique,combining with the leather defect detection algorithm based on visual saliency degree in order to improve the detection effect.First is to statistic GLCM of the image,and then calculate a kind of pixel's gray frequency distribution as its texture representation.Since the disadvantage that large amount of calculation and slow speed of the traditional texture features of pixels by GLCM such as energy,this texture expression based on GLCM which can be quickly calculated it's more suitable for applications in leather defect diction with some real-time requirement,and experiment prove that it also has slightly improved the detection effects,has a certain significance.