Automated tuning of a vision-based inspection system for industrial food manufacturing

Quality control in industrial food manufacturing can be reliably performed with computer vision systems that operate at high speed. However, most of these inspection stations need to be tuned manually and only perform well on a specific product. This research integrates machine learning techniques in the process to automate the initial tuning of real-time vision-based inspection systems for bakery products. The combination of feature selection techniques with machine learning is assessed in terms of classification performance. A formal automated tuning methodology is introduced and evaluated experimentally with data from industrial inspection stations. The work demonstrates that an inspection system automatically tuned with the proposed technique can systematically achieve 98% correct classification when compared with the classification generated with a manually tuned system.