A Deep Learning-based Method for Production Carton Packaging Defect Detection

Abstract

In the automation production packing system, appearance defects on the carton packaging films occur frequently. To solve this problem, a deep learning-based method is proposed in this work to identify film defects during the packaging process to ensure the production packaging quality. This method adopts the deformable convolution network (DCN) v2-C3 module to replace the C3 module in the Neck part of the You Only Look Once version 5 (YOLOv5s) network to extract the deep feature information of the defects in the production carton packaging, with the purpose to improve the spatial transformation ability of the detection model and the model generalization ability to different shapes of targets. Field data are used to evaluate the proposed method. The analysis results indicate that the recognition rate of the proposed method is 99.3 % for different carton packaging defects; and compared to the original YOLOv5s method, the detection accuracy of the proposed method increases by 2.7 % and the scrap rate is reduced by 1.3 %. As a result, the proposed method can meet the requirements for defect detection in the production carton packaging in practical applications.