Design of Online Detection Algorithm for Appearance Defects of Cigarette Packs Based on Deep Learning

Traditional cigarette package defect recognition devices are mostly based on the technology of image pixel template comparison. In terms of defect detection types, detection coverage areas, rapid iteration of detection parameters, and continuous upgrading of detection capabilities, they have been unable to fully meet the requirements of cigarette industry enterprises for packaging appearance quality control. The intelligent inspection algorithm of cigarette package appearance defects based on deep learning uses the feature extractor of strong and weak data enhanced contrast distribution learning and multi-scale progressive auto encoder learning, which can detect the appearance defects of product packaging efficiently, accurately and quickly. At the same time, based on the adaptive incremental network, the detection capability can be quickly migrated. This scheme has the characteristics of fast deployment, high defect detection rate, low error rejection rate, strong mobility, etc. It provides a new idea for online inspection of cigarette packaging quality, and is of great significance to reduce the risk of product appearance defects entering the market.