Research And Development For Leather Defect Inspection System Based On Machine Vision Detection Technology

Leather products as high-end luxury consumer goods,its quality is vital to the survival of enterprises.Currently,defects of leather products,such as defects,scratches,wormholes,holes,wrinkles,stains and other defects,are basically detected by artificial naked eyes,which is not only inefficient but also harmful to workers’ physical and mental health.How to use advanced machine vision technology to improve the efficiency of leather defect detection has become a top priority for leather products enterprises.The title of this paper comes from a leather products enterprise in Foshan,guangdong province.Aiming at the existing defect detection problems of leather products in this enterprise,a new leather defect detection algorithm based on machine vision based on Gabor wavelet transform is proposed.The novelty of this algorithm lies in the improvement of the largest inter-class variance for defect detection.In order to realize and verify the algorithm,the author builds the detection system platform,proposes the defect detection scheme,designs the defect detection algorithm,and realizes the defect classification algorithm.The specific research and development work of the paper is as follows:Firstly,the overall design scheme of the system and its visual detection module platform are set up.The overall scheme of the system includes the selection of its hardware equipment,work flow and overall framework structure;The visual inspection module mainly introduces the software framework and operating procedures.Secondly,this paper focuses on the leather defect detection algorithm and its classification algorithm.For the detection algorithm,through research and analysis,a defect detection algorithm based on Gabor wavelet is proposed.The algorithm detects leather defect through Gabor wavelet and divides various defect areas by the improved largest inter-class variance.Under the premise of universality and accuracy,the real-time performance of the algorithm is improved.For the defect types of classification algorithm,this paper proposes an algorithm of defect classification that based on SVM.In this algorithm,the feature values of defect categories are extracted by means of gray level co-occurrence matrix,and then the SVM model is used for training and classification judgment.For a wide variety of defects and small sample data sets,the SVM classification model have a stronger learning ability and high accuracy.Thirdly,the detection algorithm and defect classification algorithm were tested and verified through the actual data set,and the results showed that the accuracy of the detection algorithm was increased by 4.5%,while the speed and accuracy of the defect classification algorithm were increased by 3.2% and 3.6% respectively.Finally,this paper summarizes the problems found in the research stage and the future research direction.The research and development system is still in the development and test stage.Although the running speed and accuracy have been improved,its stability and data set still have shortcomings.Only continuous improvement in the later stage can meet the market demand.