Machine Vision-based Integrated Inspection System for Beverage Bottles with Multiple Defects

Abstract

Currently, in the beverage filling production line, the corresponding detection device is usually installed after each production process to detect whether the beverage is defective. This method of installing testing devices in different working positions can classify and eliminate defects of defective products, but it needs to manufacture multiple sets of different types of defective testing machines, which has problems such as high resource consumption and waste of installation space. In order to solve these problems, an expandable multi-functional integrated inspection system has been designed in this paper, which can synchronize defects detection in different processes. The system includes a multitasking detection network based on multi-class decoders. By jointly training the detection network with the collected data set, the results of beverage bottle label segmentation, region of interest (ROI) detection, and code recognition can be obtained simultaneously, thus the system can effectively realize the synchronous detection of different defects such as label and spray code missing, inversion and so on. The system also provides an expandable multitasking interface, which can be extended according to the other actual requirements of the production line. Experimental results show that the system achieves 96% detection accuracy, 18000 bottles/h detection speed and good robustness.