Button Recognition And Scratch Detection System Based On Machine Vision

Button is a kind of necessities,with a wide variety of characteristics similar to the style.Because its production and consumption are very impressive,there is a huge market and profits.The collection of button information can be established a large-scale similar sample database.Besides,button production is labor-intensive industries,requires a lot of labor,especially in terms of quality control.With the arrival of the era ofbig data,the research of classification and quality detection of buttons is based on the market and actual needs.In the large button data system,through the button product identification technology,people can quickly find the button manufacturers and manufacturers button template,this will reduce the development costs of button manufacturers,and then get more economic benefits.In terms of button recognition,a variety of button samples can be grouped according to certain standards to reduce the difficulty of identification,so this paper proposes two group strategies.The first one is a dynamic grouping,with the aim of distributing all kinds of button to each group as evenly as possible by calculating entropy.The second one is a static grouping,with the aim of reducing the probability that the sample is classified into the wrong sample group during the recognition process.And then extract the color,size and texture of each group of samples,after PCA dimensionality input to the SVM for training.Through the actual test of 118 kinds of similar buttons,the correct identify rate of the first algorithm is 86.875%;the correct identify rate of the second algorithm is 93.75%.In terms of button scratches,because the production process will produce ring texture,which will interference the detection of scratch,this paper proposed an isotropic ring texture suppression algorithm can significantly reduce the background interference.Besides,the scratch may be broken when extracting the characteristics of scratch,this paper proposed a discontinuous scratches detection algorithm.The correct rate of scratch detect is 92.5%.In order to ensure the reliability of identification and detection,all sample data are collected in the environment which simulated factory actual production.Due to button production process has the characteristics of style updates are fast and yield is huge,the system aims to reduce the sample acquisition and model training time,to quickly detect defects and reduce the cost of testing equipment,to meet the actual needs of production as maximum as possible.