A Machine Vision Apparatus and Method for Can-End Inspection

As a critical component of metal can containers, can end and its manufacture quality are closely relevant to product safety in food and beverage industry. To satisfy the requirements of quality control, a machine vision apparatus for real-time can-end inspection is presented in this paper. With a brief description of the apparatus system design and imaging system, our emphasis is put on the postprocessing image analysis. To detect defects and deformations across the imaged can-end surface, an entropy-rate clustering algorithm combined with prior shape constraint is proposed to locate the can-end object and divide it into multiple measuring regions. Then, a superpixel grouping and selection scheme is adopted to find defective areas inside the flat central panel. For the other three annular measuring regions, a multiscale ridge detection algorithm is introduced to seek defects and deformations along their projection profiles. According to in-line experiments and test, our apparatus can find out a majority of the can-end defects with a detection accuracy as high as 99.48% for various circular can ends.