Measures of similarity and detection of miniature foreign bodies in packaged foods

Foreign bodies (FB) or particles submersed in food containers constitute a hazard and quality assurance issue for many food manufacturers. Ultrasound has demonstrated potential for detecting relatively large FBs in packaged foods. In the case of miniature FBs, the ultrasound signal goes through minimal phase and amplitude change, and as a result, small FBs may end up undetected. Similarity measures were used to capture the minute changes of the ultrasound signals during FB detection. Three measures in the investigation were Minkowski, Chebychev, and Cityblock metrics. The sensitivity of the three measures was evaluated in capturing minute signal changes with numerical signals, whose changes are quantitatively controlled. The experiments were conducted to detect small steel particles in a glass container. The Minkowski metric showed numerically and experimentally better FB detection capabilities. Two signal alignment algorithms were introduced to minimize the impact of signal misalignment on similarity analysis.