Research And Application Of Two Defect Inspection Algorithms Based On Machine Vision

The demand for industrial products is getting increasingly high along with the continuous development of China's economy and the improvement of people's living standard.Industrial production capacity is also increasing rapidly as industrial automation improving continuously,which brings many difficulties to product quality inspection.In recent years,the rapid development of machine vision has provided a possibility for automatic inspection of products,and the defect inspection based on machine vision has gradually become an essential part of industrial automation.The research of defect inspection algorithms based on machine vision is very important in engineering applications.According to statistics,the total volume of mobile phone production in 2017 is around 0.45 billion,of which most of the batteries are produced in China.Artificial inspection of cellphone battery character is apt to bring high rate of over-killed and over-passed,because of high requirement of inspection efficiency and precision,and non-unified inspection standard of many kinds of batteries.Therefore,it is urgent to develop machine vision based automatic defect inspection systems.In view of that the extraction accuracy of ROI using existing detected images for battery character items is not enough,a multi-step localization algorithm based on affine transformation and template matching is proposed in this thesis.The algorithmic implementation is as follows:First,use the geometric location method to complete the coarse location of battery characters through a single location area;Secondly,use the two-point positioning method to complete the accurate location of battery characters through double location area;Finally,use the template matching method to complete the location error compensation of single character area.The two-step affine transformation is used to rotate and translate the image,which overcomes the problem that the template cannot match the rotating image,and the precision error is compensated by template matching.The algorithm combines the advantages of the three localization algorithms,such as fast computation speed and high registration accuracy.In recent years,with the rise of the takeaway industry,the market demand for degradable disposable tableware grows sharply.Because of the huge production and low artificial inspection efficiency,the disposable tableware manufacturers now have the demand for machine vision based intelligent inspection.Realizing that the defects the degradable throwaway tableware is not needed to be classified,quartering was proposed to quickly judge the defect concentration.The algorithm is a generalization of the traditional dichotomy in the two-dimensional plane.The most intensive region of the defects can be detected by iterative segmentation.It has many advantages such as fast speed,simplicity,easy programming and easy to be extended.The detection results of the two algorithms are verified by field testing and experiment.The results show that the detection speed and recognition rate are improved significantly,which can meet the requirements of industrial production.