Research On Diapers' Defects Detection Based On Machine Vision

Machine vision,with its advantages of high speed,high precision and good stability,is widely used in the automatic detection of product defects,and has become a research hotspot.Based on the surface defects(appearance defects,cracks,stains,holes,layers)of the surface structure of diapers,this paper carried out the relevant machine vision surface defect detection research,and used machine learning method to complete the defect classification.The main work accomplished in this paper is as follows:Firstly,sufficient research has been done on the current surface defect detection methods based on machine vision.Several representative detection methods are sorted out to provide ideas for this paper.Secondly,according to the actual characteristics of defect detection in this paper,the design of the experimental platform system was completed,and the required hardware was selected reasonably.The experimental platform built could well complete the task in this paper.Thirdly,according to the characteristics of various defects,the corresponding detection methods are proposed to realize the detection of various defects.In this paper,MeanShift filter is combined with the product shape extraction method to completely retain the object shape characteristics,and design the minimum external rectangle acquisition method suitable for the product shape characteristics in this paper,which simplifies the acquisition process.In this paper,the concept of reverse center is put forward,which is used to strengthen the crack defect.In this paper,an adaptive threshold segmentation method based on normal distribution confidence interval is proposed,which greatly improves the stability and accuracy of detecting blemishes and holes.Then,the defect detection method in this paper is used to complete the testing experiments for various defects,and the experimental results are analyzed.In this paper,three indicators(detection rate,omission rate and error rate)were used to evaluate the testing effect.The experimental results showed that the average detection rate of various defects was 97.48%,the average omission rate was 2.52% and the average error rate was 1.12%.Finally,the classification of multiple defects is realized by using the "one-to-one" strategy of SVM algorithm,and the experimental results of classification are analyzed,and corresponding improvement ideas are proposed.In this paper,the length,width,area and contrast of defects are used as feature data to constitute the feature vector of defect classification.The final experimental results show that the average accuracy of defect classification is 90.62%.In this paper,the theoretical exploration of machine vision detection of surface defects in diapers is completed,and a good detection effect is obtained.