Research On Quality Inspection Method Of A Class Of Industrial Assembly Products Based On Image Recognition

With the continuous improvement of automation level and the continuous development of computer technology and information technology in recent years,more and more machine vision means are introduced into the production process of enterprises.Through the use of image recognition detection system to detect various components of industrial assembly products,in order to identify whether each component is installed correctly.Machine vision,instead of manual operation,plays an important role in improving the accuracy of the detection system,reducing the complexity and cost of the system,and improving the efficiency of labor production.In this paper,a quality detection method based on image recognition is designed for the detection process of industrial assembly products such as automobile fuse box,and a large number of experiments are carried out.The work of this paper includes:(1)Select the appropriate hardware and software to build a visual inspection system,based on Windows 10,visual studio 2015,opencv3.4.10 open source visual library,Haikang camera SDK and other development environment,and design a set of hardware platform based on industrial computer,industrial camera,diffuse light source.At the same time,the common image preprocessing methods such as image filtering,binarization,edge detection and image enhancement are studied,and the image preprocessing method suitable for industrial production process is selected.(2)Image color features.Compared with the traditional color model,this paper proposes a Hg one-dimensional color model which is more suitable for the color distribution of surface devices of industrial assembly products,such as automobile fuse box,and optimizes the parameters of the model by mountain climbing method to further enhance the ability of the model to distinguish different color devices.(3)Image character features.Based on the basic image processing methods and SVM classifier,this paper proposes an automatic character location and recognition method based on SVM.Through the image data set of the characters on the surface of existing devices,the most appropriate penalty factor is selected and the SVM classifier model is trained.Then the characters in the automatically segmented character region are separated and sent to the classifier for recognition.(4)Image texture features.Compared with SIFT,surf,orb and other commonly used feature extraction methods,combined with Hamming distance and violence matching method,this paper proposes an image texture feature matching method based on orb feature extraction algorithm.Through contrast enhancement,matching point selection and other methods,the ability of the algorithm to distinguish image texture features is improved.Based on the above method,the quality inspection system of industrial assembly products can finally reach a correct detection rate over 99%,and the average single image detection time is less than 2s,which meets the basic requirements of industrial production.